{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:YYKVRPH4HEHH7MP4UCPN3JEQAK","short_pith_number":"pith:YYKVRPH4","schema_version":"1.0","canonical_sha256":"c61558bcfc390e7fb1fca09edda49002b5cd412675087975fafda933f8f823d3","source":{"kind":"arxiv","id":"2506.03933","version":2},"attestation_state":"computed","paper":{"title":"Diffusion-based Cumulative Adversarial Purification for Vision Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Anders Holst, Jia Fu, Kunyu Peng, Sepideh Pashami, Volkan Cevher, Xiao Zhang, Yihang Chen, Yongtao Wu","submitted_at":"2025-06-04T13:26:33Z","abstract_excerpt":"Vision Language Models (VLMs) have shown remarkable capabilities in multimodal understanding, yet their susceptibility to adversarial perturbations poses a significant threat to their reliability in real-world applications. Despite often being imperceptible to humans, these perturbations can drastically alter model outputs, leading to erroneous interpretations and decisions. This paper introduces DiffCAP, a novel diffusion-based purification strategy that can effectively neutralize adversarial corruptions in VLMs. We theoretically establish a provable recovery region in the forward diffusion p"},"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":"2506.03933","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-06-04T13:26:33Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"64eac57c603b3e49713d0510cfaaef347aadddf9dc33a09771ed98def8098942","abstract_canon_sha256":"978b977f4b08ab3ed2860e27524d077d963f211aa4e64c9cc7966c3c94b1394a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-11T01:09:13.848232Z","signature_b64":"fuOQ3Qvmpv4j+US0vEV4mPqfRdX7e39hdaINdiNMiRI2gOgfRSwMb1/K6goP/gYoXv42/DyACRCeyikiYc8oCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c61558bcfc390e7fb1fca09edda49002b5cd412675087975fafda933f8f823d3","last_reissued_at":"2026-06-11T01:09:13.847337Z","signature_status":"signed_v1","first_computed_at":"2026-06-11T01:09:13.847337Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Diffusion-based Cumulative Adversarial Purification for Vision Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Anders Holst, Jia Fu, Kunyu Peng, Sepideh Pashami, Volkan Cevher, Xiao Zhang, Yihang Chen, Yongtao Wu","submitted_at":"2025-06-04T13:26:33Z","abstract_excerpt":"Vision Language Models (VLMs) have shown remarkable capabilities in multimodal understanding, yet their susceptibility to adversarial perturbations poses a significant threat to their reliability in real-world applications. Despite often being imperceptible to humans, these perturbations can drastically alter model outputs, leading to erroneous interpretations and decisions. This paper introduces DiffCAP, a novel diffusion-based purification strategy that can effectively neutralize adversarial corruptions in VLMs. We theoretically establish a provable recovery region in the forward diffusion p"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.03933","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/2506.03933/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":"2506.03933","created_at":"2026-06-11T01:09:13.847459+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.03933v2","created_at":"2026-06-11T01:09:13.847459+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.03933","created_at":"2026-06-11T01:09:13.847459+00:00"},{"alias_kind":"pith_short_12","alias_value":"YYKVRPH4HEHH","created_at":"2026-06-11T01:09:13.847459+00:00"},{"alias_kind":"pith_short_16","alias_value":"YYKVRPH4HEHH7MP4","created_at":"2026-06-11T01:09:13.847459+00:00"},{"alias_kind":"pith_short_8","alias_value":"YYKVRPH4","created_at":"2026-06-11T01:09:13.847459+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2511.21893","citing_title":"Breaking the Illusion: Consensus-Based Generative Mitigation of Adversarial Illusions in Multi-Modal Embeddings","ref_index":8,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YYKVRPH4HEHH7MP4UCPN3JEQAK","json":"https://pith.science/pith/YYKVRPH4HEHH7MP4UCPN3JEQAK.json","graph_json":"https://pith.science/api/pith-number/YYKVRPH4HEHH7MP4UCPN3JEQAK/graph.json","events_json":"https://pith.science/api/pith-number/YYKVRPH4HEHH7MP4UCPN3JEQAK/events.json","paper":"https://pith.science/paper/YYKVRPH4"},"agent_actions":{"view_html":"https://pith.science/pith/YYKVRPH4HEHH7MP4UCPN3JEQAK","download_json":"https://pith.science/pith/YYKVRPH4HEHH7MP4UCPN3JEQAK.json","view_paper":"https://pith.science/paper/YYKVRPH4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.03933&json=true","fetch_graph":"https://pith.science/api/pith-number/YYKVRPH4HEHH7MP4UCPN3JEQAK/graph.json","fetch_events":"https://pith.science/api/pith-number/YYKVRPH4HEHH7MP4UCPN3JEQAK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YYKVRPH4HEHH7MP4UCPN3JEQAK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YYKVRPH4HEHH7MP4UCPN3JEQAK/action/storage_attestation","attest_author":"https://pith.science/pith/YYKVRPH4HEHH7MP4UCPN3JEQAK/action/author_attestation","sign_citation":"https://pith.science/pith/YYKVRPH4HEHH7MP4UCPN3JEQAK/action/citation_signature","submit_replication":"https://pith.science/pith/YYKVRPH4HEHH7MP4UCPN3JEQAK/action/replication_record"}},"created_at":"2026-06-11T01:09:13.847459+00:00","updated_at":"2026-06-11T01:09:13.847459+00:00"}