vimarsh's picture
Update logbook: Reproduction: Full-Batch Gradient Descent Outperforms One-Pass SGD: Sample Complexity Separation in Single-Index Learning
b5e052f verified
Raw History Blame Contribute Delete
3.88 kB
{
"schema_version": 2,
"title": "Reproduction: Full-Batch Gradient Descent Outperforms One-Pass SGD: Sample Complexity Separation in Single-Index Learning",
"emoji": "🎯",
"space_id": "vimarsh/repro-full-batch-gd-outperforms-one-pass-sgd-sample-complexity-separation",
"paper": {
"arxiv_id": "2602.02431"
},
"tags": [
"icml2026-repro",
"paper-QItZDBVCT0"
],
"updated_at": "2026-07-26T18:32:48+00:00",
"root": {
"slug": "index",
"title": "Reproduction: Full-Batch Gradient Descent Outperforms One-Pass SGD: Sample Complexity Separation in Single-Index Learning",
"file": "pages/index.md",
"children": [
{
"slug": "executive-summary",
"title": "Executive summary",
"file": "pages/executive-summary/page.md",
"children": []
},
{
"slug": "claim-1-quadratic-activation-no-full-batch-gain-n-o-d-log-d-fails-thm-3-1",
"title": "Claim 1: Quadratic activation: no full-batch gain, n = o(d log d) fails (Thm 3.1)",
"file": "pages/claim-1-quadratic-activation-no-full-batch-gain-n-o-d-log-d-fails-thm-3-1/page.md",
"children": []
},
{
"slug": "claim-2-truncated-activation-full-batch-weak-recovery-at-n-d-thm-3-2",
"title": "Claim 2: Truncated activation: full-batch weak recovery at n ≳ d (Thm 3.2)",
"file": "pages/claim-2-truncated-activation-full-batch-weak-recovery-at-n-d-thm-3-2/page.md",
"children": []
},
{
"slug": "claim-3-strong-recovery-in-t-log-d-steps-with-n-cm-d-thm-4-1",
"title": "Claim 3: Strong recovery in T ≳ log d steps with n ≥ CM⁴d (Thm 4.1)",
"file": "pages/claim-3-strong-recovery-in-t-log-d-steps-with-n-cm-d-thm-4-1/page.md",
"children": []
},
{
"slug": "claim-4-two-phase-gd-trajectory-angle-reduction-then-geometric-refinement-sec-4",
"title": "Claim 4: Two-phase GD trajectory: angle reduction then geometric refinement (Sec 4)",
"file": "pages/claim-4-two-phase-gd-trajectory-angle-reduction-then-geometric-refinement-sec-4/page.md",
"children": []
},
{
"slug": "claim-5-full-batch-matches-the-n-d-information-theoretic-limit-thm-3-2",
"title": "Claim 5: Full-batch matches the n ≳ d information-theoretic limit (Thm 3.2)",
"file": "pages/claim-5-full-batch-matches-the-n-d-information-theoretic-limit-thm-3-2/page.md",
"children": []
},
{
"slug": "conclusion",
"title": "Conclusion",
"file": "pages/conclusion/page.md",
"children": []
}
]
},
"traces": [],
"workspace": {
"file": "workspace.json",
"file_count": 0,
"total_size": 0,
"bucket_id": null
},
"agent_view_tokens": 14890,
"trace_view_tokens": 38124,
"workspace_view_tokens": 868,
"revision": "beaf720c606b02e70cf7",
"traces_ref": {
"repo_id": "vimarsh/repro-full-batch-gd-outperforms-one-pass-sgd-sample-complexity-separation-traces",
"repo_type": "dataset",
"repo_url": "https://huggingface.co/datasets/vimarsh/repro-full-batch-gd-outperforms-one-pass-sgd-sample-complexity-separation-traces",
"private": true,
"viewer_path": "trackio/index.json"
},
"trace_dataset": "https://huggingface.co/datasets/vimarsh/repro-full-batch-gd-outperforms-one-pass-sgd-sample-complexity-separation-traces",
"workspace_ref": {
"repo_id": "vimarsh/repro-full-batch-gd-outperforms-one-pass-sgd-sample-complexity-separation-artifacts",
"repo_type": "bucket",
"repo_url": "https://huggingface.co/buckets/vimarsh/repro-full-batch-gd-outperforms-one-pass-sgd-sample-complexity-separation-artifacts",
"private": true
},
"workspace_bucket": "https://huggingface.co/buckets/vimarsh/repro-full-batch-gd-outperforms-one-pass-sgd-sample-complexity-separation-artifacts"
}