{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:WUOU3S5RR32UPWWIINVGV67XPO","short_pith_number":"pith:WUOU3S5R","schema_version":"1.0","canonical_sha256":"b51d4dcbb18ef547dac8436a6afbf77bb21c2565145e9fd291fa2468604021a5","source":{"kind":"arxiv","id":"2606.05843","version":1},"attestation_state":"computed","paper":{"title":"Mechanistic Insights into Functional Sparsity in Multimodal LLMs via CoRe Heads","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Juntao Li, Min Zhang, Quantong Qiu, Ruoxi Sun, Yihang Lou, Zecheng Tang","submitted_at":"2026-06-04T08:18:31Z","abstract_excerpt":"While Multimodal Large Language Models (MLLMs) demonstrate remarkable proficiency on complex vision-language tasks, the mechanisms by which they extract query-relevant visual features from complex, noisy contexts remain opaque. In this paper, we present an in-depth interpretability study that uncovers a profound structural property within MLLMs: functional sparsity in cross-modal retrieval. Leveraging a token-level metric termed Retrieval Attention Mass (RAM), we identify and characterize a highly specialized subset of attention heads, referred to as Context-aware Retrieval (CoRe) heads. Acros"},"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":"2606.05843","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-06-04T08:18:31Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"7a054373bc4f8fd4e0b6c757993a00cd53712ace35cec26623762eaed4b9b7f7","abstract_canon_sha256":"c7e195534cc4054c16bbf0ec4bd35e2ea6bc09d69be47e809cb3f211ebdfb7a8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-05T01:15:05.408828Z","signature_b64":"yRKwbN7HKe9N3P09pMnavr2JZk6wiOvlSBGyJlF6VK0d5DhPgbCeZMZSHCWgLjdnukP7HeEi61yicl8fFu8NBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b51d4dcbb18ef547dac8436a6afbf77bb21c2565145e9fd291fa2468604021a5","last_reissued_at":"2026-06-05T01:15:05.408417Z","signature_status":"signed_v1","first_computed_at":"2026-06-05T01:15:05.408417Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mechanistic Insights into Functional Sparsity in Multimodal LLMs via CoRe Heads","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Juntao Li, Min Zhang, Quantong Qiu, Ruoxi Sun, Yihang Lou, Zecheng Tang","submitted_at":"2026-06-04T08:18:31Z","abstract_excerpt":"While Multimodal Large Language Models (MLLMs) demonstrate remarkable proficiency on complex vision-language tasks, the mechanisms by which they extract query-relevant visual features from complex, noisy contexts remain opaque. In this paper, we present an in-depth interpretability study that uncovers a profound structural property within MLLMs: functional sparsity in cross-modal retrieval. Leveraging a token-level metric termed Retrieval Attention Mass (RAM), we identify and characterize a highly specialized subset of attention heads, referred to as Context-aware Retrieval (CoRe) heads. Acros"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.05843","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/2606.05843/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":"2606.05843","created_at":"2026-06-05T01:15:05.408479+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.05843v1","created_at":"2026-06-05T01:15:05.408479+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.05843","created_at":"2026-06-05T01:15:05.408479+00:00"},{"alias_kind":"pith_short_12","alias_value":"WUOU3S5RR32U","created_at":"2026-06-05T01:15:05.408479+00:00"},{"alias_kind":"pith_short_16","alias_value":"WUOU3S5RR32UPWWI","created_at":"2026-06-05T01:15:05.408479+00:00"},{"alias_kind":"pith_short_8","alias_value":"WUOU3S5R","created_at":"2026-06-05T01:15:05.408479+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/WUOU3S5RR32UPWWIINVGV67XPO","json":"https://pith.science/pith/WUOU3S5RR32UPWWIINVGV67XPO.json","graph_json":"https://pith.science/api/pith-number/WUOU3S5RR32UPWWIINVGV67XPO/graph.json","events_json":"https://pith.science/api/pith-number/WUOU3S5RR32UPWWIINVGV67XPO/events.json","paper":"https://pith.science/paper/WUOU3S5R"},"agent_actions":{"view_html":"https://pith.science/pith/WUOU3S5RR32UPWWIINVGV67XPO","download_json":"https://pith.science/pith/WUOU3S5RR32UPWWIINVGV67XPO.json","view_paper":"https://pith.science/paper/WUOU3S5R","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.05843&json=true","fetch_graph":"https://pith.science/api/pith-number/WUOU3S5RR32UPWWIINVGV67XPO/graph.json","fetch_events":"https://pith.science/api/pith-number/WUOU3S5RR32UPWWIINVGV67XPO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WUOU3S5RR32UPWWIINVGV67XPO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WUOU3S5RR32UPWWIINVGV67XPO/action/storage_attestation","attest_author":"https://pith.science/pith/WUOU3S5RR32UPWWIINVGV67XPO/action/author_attestation","sign_citation":"https://pith.science/pith/WUOU3S5RR32UPWWIINVGV67XPO/action/citation_signature","submit_replication":"https://pith.science/pith/WUOU3S5RR32UPWWIINVGV67XPO/action/replication_record"}},"created_at":"2026-06-05T01:15:05.408479+00:00","updated_at":"2026-06-05T01:15:05.408479+00:00"}