File size: 3,506 Bytes
4a6bf4b 5cb8fd0 4a6bf4b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 | {
"model": "intraspannet",
"version": "0.1.0",
"task": "security product identification with open-set rejection",
"input": {
"flow_bytes": 1600,
"packets": 5,
"bytes_per_packet": 320,
"header_bytes": 80,
"payload_bytes": 240,
"ip_masking": true
},
"classes": [
"windows_defender",
"clamav_freshclam_win",
"clamav_freshclam",
"velociraptor",
"velociraptor_win",
"osquery",
"wazuh_agent",
"wireguard",
"openvpn",
"stunnel",
"squid_proxy",
"ntopng",
"guacamole_mgmt",
"waf_modsecurity",
"suricata_update",
"web_http",
"ssh",
"ftp"
],
"cls": [
"b_s0",
"a_lda",
"b_lda",
"a_s0"
],
"gate": [
"a_lda",
"b_lda"
],
"threshold": 0.1041,
"threshold_note": "Calibrated on the dev half of the validation split; not tuned on the test half.",
"metrics": {
"test_flow_count": 503,
"all_val_flow_count": 1006,
"product_accuracy_test": 0.8787,
"product_accuracy_all_val": 0.8817,
"tier_accuracy_test": 0.9245,
"tier_accuracy_all_val": 0.9334,
"macro_f1_supported_test": 0.7621,
"macro_f1_supported_all_val": 0.824,
"gate_auroc_test": 0.9784,
"gate_auroc_dev": 0.982,
"malicious_macro_f1_dedup_ustc": 0.9702,
"unknown_rejection_demo": 1
},
"hardware": {
"inference": "NVIDIA GPU with CUDA",
"reference_gpu": "RTX 4090 24GB",
"minimum_vram_gb": 4,
"recommended_vram_gb": 8,
"cpu_inference": false,
"cpu_note": "The GDN backbone uses Triton kernels. CPU fallback is not supported in v0.1.0.",
"single_arm_throughput_flows_s": 1017,
"four_arm_throughput_flows_s": 253,
"latency_ms_bs1": {
"single_arm": 10.7,
"four_arm": 43.1
}
},
"training": {
"pretraining": "contrastive self-supervised on 33,184 training-only flows",
"sft": "multi-task supervised fine-tuning, 60 epochs",
"rl": "REINFORCE with reward shaping, 300 episodes",
"seeds": [
0,
1,
2
]
},
"license": "Apache-2.0",
"citation": "See CITATION.cff in the code repository.",
"arms": {
"A_S0": {
"run": "A",
"variant": "standard multi-task",
"role": "classifier",
"checkpoint": "stage3_a_s0.pth",
"prototypes": "prototypes_a_s0_k3.pth"
},
"A_LDA": {
"run": "A",
"variant": "logit adjustment by class prior",
"role": "classifier and gate",
"checkpoint": "stage3_a_lda.pth",
"prototypes": "prototypes_a_lda_k3.pth"
},
"B_LDA": {
"run": "B",
"variant": "logit adjustment by class prior",
"role": "classifier and gate",
"checkpoint": "stage3_b_lda.pth",
"prototypes": "prototypes_b_lda_k3.pth"
},
"B_S0": {
"run": "B",
"variant": "standard multi-task",
"role": "classifier",
"checkpoint": "stage3_b_s0.pth",
"prototypes": "prototypes_b_s0_k3.pth"
},
"A_HN": {
"run": "A",
"variant": "hard-negative weighting",
"role": "dropped",
"checkpoint": "stage3_a_hn.pth",
"prototypes": "prototypes_a_hn_k3.pth"
},
"B_HN": {
"run": "B",
"variant": "hard-negative weighting",
"role": "dropped",
"checkpoint": "stage3_b_hn.pth",
"prototypes": "prototypes_b_hn_k3.pth"
}
},
"arm_naming": "Letter = pre-training run (A or B). Suffix = training variant: s0 standard multi-task objective, lda logit adjustment by class prior, hn hard-negative weighting."
}
|