{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:GJSDYKC365RUHSOBFQBSXXOIQF","short_pith_number":"pith:GJSDYKC3","schema_version":"1.0","canonical_sha256":"32643c285bf76343c9c12c032bddc88170b08013cd9354e9a6136840cefb4fea","source":{"kind":"arxiv","id":"2501.09446","version":2},"attestation_state":"computed","paper":{"title":"Double Visual Defense: Adversarial Pre-training and Instruction Tuning for Improving Vision-Language Model Robustness","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bhavya Kailkhura, Brian Bartoldson, Cihang Xie, Zeyu Wang","submitted_at":"2025-01-16T10:20:48Z","abstract_excerpt":"This paper investigates the robustness of vision-language models against adversarial visual perturbations and introduces a novel ``double visual defense\" to enhance this robustness. Unlike previous approaches that resort to lightweight adversarial fine-tuning of a pre-trained CLIP model, we perform large-scale adversarial vision-language pre-training from scratch using web-scale data. We then strengthen the defense by incorporating adversarial visual instruction tuning. The resulting models from each stage, $\\Delta$CLIP and $\\Delta^2$LLaVA, show substantially enhanced zero-shot robustness and "},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2501.09446","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2025-01-16T10:20:48Z","cross_cats_sorted":[],"title_canon_sha256":"b8ddd1568ed45ca6802dd9e3ea02424fbbb217594b6e593bec4003d0de1bfa59","abstract_canon_sha256":"6cea70636da2f5080e3e9f6be231c60c6ce82a77a6e2f5fb4d2e2596415f56f4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:45:38.899674Z","signature_b64":"mH0B/gw44ZqQB3mgAME8+LloxkbnBEhO7owMr2okHRJRe+elnDFwTFQV978y7/B1T6yL0XOBmsyNBgC7A6L9Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"32643c285bf76343c9c12c032bddc88170b08013cd9354e9a6136840cefb4fea","last_reissued_at":"2026-07-05T10:45:38.899181Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:45:38.899181Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Double Visual Defense: Adversarial Pre-training and Instruction Tuning for Improving Vision-Language Model Robustness","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bhavya Kailkhura, Brian Bartoldson, Cihang Xie, Zeyu Wang","submitted_at":"2025-01-16T10:20:48Z","abstract_excerpt":"This paper investigates the robustness of vision-language models against adversarial visual perturbations and introduces a novel ``double visual defense\" to enhance this robustness. Unlike previous approaches that resort to lightweight adversarial fine-tuning of a pre-trained CLIP model, we perform large-scale adversarial vision-language pre-training from scratch using web-scale data. We then strengthen the defense by incorporating adversarial visual instruction tuning. The resulting models from each stage, $\\Delta$CLIP and $\\Delta^2$LLaVA, show substantially enhanced zero-shot robustness and "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.09446","kind":"arxiv","version":2},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2501.09446/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2501.09446","created_at":"2026-07-05T10:45:38.899234+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.09446v2","created_at":"2026-07-05T10:45:38.899234+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.09446","created_at":"2026-07-05T10:45:38.899234+00:00"},{"alias_kind":"pith_short_12","alias_value":"GJSDYKC365RU","created_at":"2026-07-05T10:45:38.899234+00:00"},{"alias_kind":"pith_short_16","alias_value":"GJSDYKC365RUHSOB","created_at":"2026-07-05T10:45:38.899234+00:00"},{"alias_kind":"pith_short_8","alias_value":"GJSDYKC3","created_at":"2026-07-05T10:45:38.899234+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.03730","citing_title":"Beyond False Stability: High-Noise Drift Gating for Test-Time Adversarial Defenses in Vision-Language Models","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03713","citing_title":"Investigating Adversarial Robustness of Multi-modal Large Language Models","ref_index":62,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GJSDYKC365RUHSOBFQBSXXOIQF","json":"https://pith.science/pith/GJSDYKC365RUHSOBFQBSXXOIQF.json","graph_json":"https://pith.science/api/pith-number/GJSDYKC365RUHSOBFQBSXXOIQF/graph.json","events_json":"https://pith.science/api/pith-number/GJSDYKC365RUHSOBFQBSXXOIQF/events.json","paper":"https://pith.science/paper/GJSDYKC3"},"agent_actions":{"view_html":"https://pith.science/pith/GJSDYKC365RUHSOBFQBSXXOIQF","download_json":"https://pith.science/pith/GJSDYKC365RUHSOBFQBSXXOIQF.json","view_paper":"https://pith.science/paper/GJSDYKC3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.09446&json=true","fetch_graph":"https://pith.science/api/pith-number/GJSDYKC365RUHSOBFQBSXXOIQF/graph.json","fetch_events":"https://pith.science/api/pith-number/GJSDYKC365RUHSOBFQBSXXOIQF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GJSDYKC365RUHSOBFQBSXXOIQF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GJSDYKC365RUHSOBFQBSXXOIQF/action/storage_attestation","attest_author":"https://pith.science/pith/GJSDYKC365RUHSOBFQBSXXOIQF/action/author_attestation","sign_citation":"https://pith.science/pith/GJSDYKC365RUHSOBFQBSXXOIQF/action/citation_signature","submit_replication":"https://pith.science/pith/GJSDYKC365RUHSOBFQBSXXOIQF/action/replication_record"}},"created_at":"2026-07-05T10:45:38.899234+00:00","updated_at":"2026-07-05T10:45:38.899234+00:00"}