{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:EQBSGNN64PTG3I3LUW4G6YNIF7","short_pith_number":"pith:EQBSGNN6","schema_version":"1.0","canonical_sha256":"24032335bee3e66da36ba5b86f61a82fe6c48da5b164bac576886026f0bd3b1e","source":{"kind":"arxiv","id":"2505.22839","version":1},"attestation_state":"computed","paper":{"title":"How Do Diffusion Models Improve Adversarial Robustness?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Liu Yuezhang, Xue-Xin Wei","submitted_at":"2025-05-28T20:19:21Z","abstract_excerpt":"Recent findings suggest that diffusion models significantly enhance empirical adversarial robustness. While some intuitive explanations have been proposed, the precise mechanisms underlying these improvements remain unclear. In this work, we systematically investigate how and how well diffusion models improve adversarial robustness. First, we observe that diffusion models intriguingly increase, rather than decrease, the $\\ell_p$ distance to clean samples--challenging the intuition that purification denoises inputs closer to the original data. Second, we find that the purified images are heavil"},"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":"2505.22839","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-28T20:19:21Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"07c4b3515f567d2cd1bb5a34983eea28c735ad207abaf603b7b26ba5132e933e","abstract_canon_sha256":"1008f8c01354359b136b7cf6365dc20ca9757b14e07da0b2d23fcbf27b70f0f4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:11:50.656009Z","signature_b64":"wMghCp9VeSUVUUm5uuJLWm530qWooPrJu59szG32JOKYvai2w/UEx6ar5fzwUhmV9gxjRPSF/BnrkxqYieoKCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"24032335bee3e66da36ba5b86f61a82fe6c48da5b164bac576886026f0bd3b1e","last_reissued_at":"2026-07-05T11:11:50.655419Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:11:50.655419Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How Do Diffusion Models Improve Adversarial Robustness?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Liu Yuezhang, Xue-Xin Wei","submitted_at":"2025-05-28T20:19:21Z","abstract_excerpt":"Recent findings suggest that diffusion models significantly enhance empirical adversarial robustness. While some intuitive explanations have been proposed, the precise mechanisms underlying these improvements remain unclear. In this work, we systematically investigate how and how well diffusion models improve adversarial robustness. First, we observe that diffusion models intriguingly increase, rather than decrease, the $\\ell_p$ distance to clean samples--challenging the intuition that purification denoises inputs closer to the original data. Second, we find that the purified images are heavil"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.22839","kind":"arxiv","version":1},"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/2505.22839/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":"2505.22839","created_at":"2026-07-05T11:11:50.655505+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.22839v1","created_at":"2026-07-05T11:11:50.655505+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.22839","created_at":"2026-07-05T11:11:50.655505+00:00"},{"alias_kind":"pith_short_12","alias_value":"EQBSGNN64PTG","created_at":"2026-07-05T11:11:50.655505+00:00"},{"alias_kind":"pith_short_16","alias_value":"EQBSGNN64PTG3I3L","created_at":"2026-07-05T11:11:50.655505+00:00"},{"alias_kind":"pith_short_8","alias_value":"EQBSGNN6","created_at":"2026-07-05T11:11:50.655505+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EQBSGNN64PTG3I3LUW4G6YNIF7","json":"https://pith.science/pith/EQBSGNN64PTG3I3LUW4G6YNIF7.json","graph_json":"https://pith.science/api/pith-number/EQBSGNN64PTG3I3LUW4G6YNIF7/graph.json","events_json":"https://pith.science/api/pith-number/EQBSGNN64PTG3I3LUW4G6YNIF7/events.json","paper":"https://pith.science/paper/EQBSGNN6"},"agent_actions":{"view_html":"https://pith.science/pith/EQBSGNN64PTG3I3LUW4G6YNIF7","download_json":"https://pith.science/pith/EQBSGNN64PTG3I3LUW4G6YNIF7.json","view_paper":"https://pith.science/paper/EQBSGNN6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.22839&json=true","fetch_graph":"https://pith.science/api/pith-number/EQBSGNN64PTG3I3LUW4G6YNIF7/graph.json","fetch_events":"https://pith.science/api/pith-number/EQBSGNN64PTG3I3LUW4G6YNIF7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EQBSGNN64PTG3I3LUW4G6YNIF7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EQBSGNN64PTG3I3LUW4G6YNIF7/action/storage_attestation","attest_author":"https://pith.science/pith/EQBSGNN64PTG3I3LUW4G6YNIF7/action/author_attestation","sign_citation":"https://pith.science/pith/EQBSGNN64PTG3I3LUW4G6YNIF7/action/citation_signature","submit_replication":"https://pith.science/pith/EQBSGNN64PTG3I3LUW4G6YNIF7/action/replication_record"}},"created_at":"2026-07-05T11:11:50.655505+00:00","updated_at":"2026-07-05T11:11:50.655505+00:00"}