{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ID67LVCXRBB52RNNTYSRR3ZWCB","short_pith_number":"pith:ID67LVCX","schema_version":"1.0","canonical_sha256":"40fdf5d4578843dd45ad9e2518ef361056bd3a56db60824b52ad8edbfa691137","source":{"kind":"arxiv","id":"2410.07149","version":2},"attestation_state":"computed","paper":{"title":"Towards Interpreting Visual Information Processing in Vision-Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Clement Neo, David Krueger, Fazl Barez, Luke Ong, Mor Geva, Philip Torr","submitted_at":"2024-10-09T17:55:02Z","abstract_excerpt":"Vision-Language Models (VLMs) are powerful tools for processing and understanding text and images. We study the processing of visual tokens in the language model component of LLaVA, a prominent VLM. Our approach focuses on analyzing the localization of object information, the evolution of visual token representations across layers, and the mechanism of integrating visual information for predictions. Through ablation studies, we demonstrated that object identification accuracy drops by over 70\\% when object-specific tokens are removed. We observed that visual token representations become increa"},"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":"2410.07149","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2024-10-09T17:55:02Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"b0891b1cd4a67f7ee725829ad7b87ebe64785a7d328dbe4af76d3e713ae84f67","abstract_canon_sha256":"84fa500758c105670ef69ffc7eed09a75393f549d9cfc9d4b5c0596fe0e33930"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:54:13.375304Z","signature_b64":"53yReb7pC7tqFtOo1fXux70P4NZKuUPoxTeLM+jdF53MmJ/IQLVuAwjN7YBYJIunJNcpPC9Pv00LNN9x6OSEDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"40fdf5d4578843dd45ad9e2518ef361056bd3a56db60824b52ad8edbfa691137","last_reissued_at":"2026-07-05T10:54:13.374800Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:54:13.374800Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards Interpreting Visual Information Processing in Vision-Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Clement Neo, David Krueger, Fazl Barez, Luke Ong, Mor Geva, Philip Torr","submitted_at":"2024-10-09T17:55:02Z","abstract_excerpt":"Vision-Language Models (VLMs) are powerful tools for processing and understanding text and images. We study the processing of visual tokens in the language model component of LLaVA, a prominent VLM. Our approach focuses on analyzing the localization of object information, the evolution of visual token representations across layers, and the mechanism of integrating visual information for predictions. Through ablation studies, we demonstrated that object identification accuracy drops by over 70\\% when object-specific tokens are removed. We observed that visual token representations become increa"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.07149","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/2410.07149/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":"2410.07149","created_at":"2026-07-05T10:54:13.374861+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.07149v2","created_at":"2026-07-05T10:54:13.374861+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.07149","created_at":"2026-07-05T10:54:13.374861+00:00"},{"alias_kind":"pith_short_12","alias_value":"ID67LVCXRBB5","created_at":"2026-07-05T10:54:13.374861+00:00"},{"alias_kind":"pith_short_16","alias_value":"ID67LVCXRBB52RNN","created_at":"2026-07-05T10:54:13.374861+00:00"},{"alias_kind":"pith_short_8","alias_value":"ID67LVCX","created_at":"2026-07-05T10:54:13.374861+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":16,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06445","citing_title":"Analysis-by-Proxy: Localization Signals in VLMs Operating as Condition Encoders","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2606.20077","citing_title":"The Hidden Evolution of Disguised Visual Context inside the VLM","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2606.10147","citing_title":"From Senses to Decisions: The Information Flow of Auditory and Visual Perception in Multimodal LLMs","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28273","citing_title":"Vision-Default, Prior-Override: Causal Mechanisms of Perception-Knowledge Conflict in Vision-Language Models","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08245","citing_title":"When Language Overwrites Vision: Over-Alignment and Geometric Debiasing in Vision-Language Models","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2511.14159","citing_title":"MVI-Bench: A Comprehensive Benchmark for Evaluating Robustness to Misleading Visual Inputs in LVLMs","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08245","citing_title":"When Language Overwrites Vision: Over-Alignment and Geometric Debiasing in Vision-Language Models","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2509.14837","citing_title":"V-SEAM: Visual Semantic Editing and Attention Modulating for Causal Interpretability of Vision-Language Models","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2511.17699","citing_title":"Understanding Counting Mechanisms in Large Language and Vision-Language Models","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08245","citing_title":"When Language Overwrites Vision: Over-Alignment and Geometric Debiasing in Vision-Language Models","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2604.02486","citing_title":"VLMs Need Words: Vision Language Models Ignore Visual Detail In Favor of Semantic Anchors","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03045","citing_title":"STEAR: Layer-Aware Spatiotemporal Evidence Intervention for Hallucination Mitigation in Video Large Language Models","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08245","citing_title":"When Language Overwrites Vision: Over-Alignment and Geometric Debiasing in Vision-Language Models","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19083","citing_title":"ProjLens: Unveiling the Role of Projectors in Multimodal Model Safety","ref_index":168,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17941","citing_title":"From Heads to Neurons: Causal Attribution and Steering in Multi-Task Vision-Language Models","ref_index":92,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04641","citing_title":"CAST: Mitigating Object Hallucination in Large Vision-Language Models via Caption-Guided Visual Attention Steering","ref_index":26,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ID67LVCXRBB52RNNTYSRR3ZWCB","json":"https://pith.science/pith/ID67LVCXRBB52RNNTYSRR3ZWCB.json","graph_json":"https://pith.science/api/pith-number/ID67LVCXRBB52RNNTYSRR3ZWCB/graph.json","events_json":"https://pith.science/api/pith-number/ID67LVCXRBB52RNNTYSRR3ZWCB/events.json","paper":"https://pith.science/paper/ID67LVCX"},"agent_actions":{"view_html":"https://pith.science/pith/ID67LVCXRBB52RNNTYSRR3ZWCB","download_json":"https://pith.science/pith/ID67LVCXRBB52RNNTYSRR3ZWCB.json","view_paper":"https://pith.science/paper/ID67LVCX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.07149&json=true","fetch_graph":"https://pith.science/api/pith-number/ID67LVCXRBB52RNNTYSRR3ZWCB/graph.json","fetch_events":"https://pith.science/api/pith-number/ID67LVCXRBB52RNNTYSRR3ZWCB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ID67LVCXRBB52RNNTYSRR3ZWCB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ID67LVCXRBB52RNNTYSRR3ZWCB/action/storage_attestation","attest_author":"https://pith.science/pith/ID67LVCXRBB52RNNTYSRR3ZWCB/action/author_attestation","sign_citation":"https://pith.science/pith/ID67LVCXRBB52RNNTYSRR3ZWCB/action/citation_signature","submit_replication":"https://pith.science/pith/ID67LVCXRBB52RNNTYSRR3ZWCB/action/replication_record"}},"created_at":"2026-07-05T10:54:13.374861+00:00","updated_at":"2026-07-05T10:54:13.374861+00:00"}