{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:J2N4UVB76ER7TS76MMADLMSGKR","short_pith_number":"pith:J2N4UVB7","schema_version":"1.0","canonical_sha256":"4e9bca543ff123f9cbfe630035b2465458239316375c4bb426c4abbfdb6e86fb","source":{"kind":"arxiv","id":"2601.21531","version":2},"attestation_state":"computed","paper":{"title":"On the Adversarial Robustness of Large Vision-Language Models under Visual Token Compression","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.CR","authors_text":"Haibo Hu, Hangcheng Liu, Hao Wang, Li Bai, Qingqing Ye, Tianwei Zhang, Xinwei Zhang","submitted_at":"2026-01-29T10:47:21Z","abstract_excerpt":"Visual token compression is widely used to accelerate large vision-language models (LVLMs) by pruning or merging visual tokens, yet its adversarial robustness remains unexplored. We show that existing encoder-based attacks cannot fully disclose the robustness vulnerabilities of compressed LVLMs, due to an optimization-inference mismatch: perturbations are optimized on the full-token representation, while inference is performed through a token-compression bottleneck. To address this gap, we propose the Compression-AliGnEd attack (CAGE), which aligns perturbation optimization with compression in"},"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":"2601.21531","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CR","submitted_at":"2026-01-29T10:47:21Z","cross_cats_sorted":["cs.AI","cs.CV"],"title_canon_sha256":"2acca2860c2299ffffcbdbb4b86ad2f9c83f7f36e3e62a49ca5c52a47c09e195","abstract_canon_sha256":"0837c565843940325f41ccb290ca55adfa0e066a064161293374d985161fb3cc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:03:03.567780Z","signature_b64":"kDuAeAf2z0+EFHC+cuPab8rAnHxXpzV3kIxSwHIyFDsPM9vltXnrgyCXsoonWchjZq/ryKnByV8zItdMeewLCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4e9bca543ff123f9cbfe630035b2465458239316375c4bb426c4abbfdb6e86fb","last_reissued_at":"2026-05-20T00:03:03.566928Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:03:03.566928Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On the Adversarial Robustness of Large Vision-Language Models under Visual Token Compression","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.CR","authors_text":"Haibo Hu, Hangcheng Liu, Hao Wang, Li Bai, Qingqing Ye, Tianwei Zhang, Xinwei Zhang","submitted_at":"2026-01-29T10:47:21Z","abstract_excerpt":"Visual token compression is widely used to accelerate large vision-language models (LVLMs) by pruning or merging visual tokens, yet its adversarial robustness remains unexplored. We show that existing encoder-based attacks cannot fully disclose the robustness vulnerabilities of compressed LVLMs, due to an optimization-inference mismatch: perturbations are optimized on the full-token representation, while inference is performed through a token-compression bottleneck. To address this gap, we propose the Compression-AliGnEd attack (CAGE), which aligns perturbation optimization with compression in"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2601.21531","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/2601.21531/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":"2601.21531","created_at":"2026-05-20T00:03:03.567066+00:00"},{"alias_kind":"arxiv_version","alias_value":"2601.21531v2","created_at":"2026-05-20T00:03:03.567066+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.21531","created_at":"2026-05-20T00:03:03.567066+00:00"},{"alias_kind":"pith_short_12","alias_value":"J2N4UVB76ER7","created_at":"2026-05-20T00:03:03.567066+00:00"},{"alias_kind":"pith_short_16","alias_value":"J2N4UVB76ER7TS76","created_at":"2026-05-20T00:03:03.567066+00:00"},{"alias_kind":"pith_short_8","alias_value":"J2N4UVB7","created_at":"2026-05-20T00:03:03.567066+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/J2N4UVB76ER7TS76MMADLMSGKR","json":"https://pith.science/pith/J2N4UVB76ER7TS76MMADLMSGKR.json","graph_json":"https://pith.science/api/pith-number/J2N4UVB76ER7TS76MMADLMSGKR/graph.json","events_json":"https://pith.science/api/pith-number/J2N4UVB76ER7TS76MMADLMSGKR/events.json","paper":"https://pith.science/paper/J2N4UVB7"},"agent_actions":{"view_html":"https://pith.science/pith/J2N4UVB76ER7TS76MMADLMSGKR","download_json":"https://pith.science/pith/J2N4UVB76ER7TS76MMADLMSGKR.json","view_paper":"https://pith.science/paper/J2N4UVB7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2601.21531&json=true","fetch_graph":"https://pith.science/api/pith-number/J2N4UVB76ER7TS76MMADLMSGKR/graph.json","fetch_events":"https://pith.science/api/pith-number/J2N4UVB76ER7TS76MMADLMSGKR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/J2N4UVB76ER7TS76MMADLMSGKR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/J2N4UVB76ER7TS76MMADLMSGKR/action/storage_attestation","attest_author":"https://pith.science/pith/J2N4UVB76ER7TS76MMADLMSGKR/action/author_attestation","sign_citation":"https://pith.science/pith/J2N4UVB76ER7TS76MMADLMSGKR/action/citation_signature","submit_replication":"https://pith.science/pith/J2N4UVB76ER7TS76MMADLMSGKR/action/replication_record"}},"created_at":"2026-05-20T00:03:03.567066+00:00","updated_at":"2026-05-20T00:03:03.567066+00:00"}