{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:YW7HCI7L4XCBMTMRUAWPZBBFSU","short_pith_number":"pith:YW7HCI7L","schema_version":"1.0","canonical_sha256":"c5be7123ebe5c4164d91a02cfc8425950d94b57ab4e0694c13d9e4ab40d13c0e","source":{"kind":"arxiv","id":"2410.02247","version":3},"attestation_state":"computed","paper":{"title":"Theoretical Insights into Fine-Tuning Attention Mechanism: Generalization and Optimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Bin Wang, Gengze Xu, Hongjin Qian, Jian Luan, Wei Liu, Xiaolin Hu, Xinhao Yao, Yong Liu","submitted_at":"2024-10-03T06:37:37Z","abstract_excerpt":"Large Language Models (LLMs), built on Transformer architectures, exhibit remarkable generalization across a wide range of tasks. However, fine-tuning these models for specific tasks remains resource-intensive due to their extensive parameterization. In this paper, we explore two remarkable phenomena related to the attention mechanism during the fine-tuning of LLMs (where $\\mathbf{W}_q$, $\\mathbf{W}_k$, and $\\mathbf{W}_v$ denote the weights of the query, key, and value layers, respectively). The first phenomenon, termed \"Unequal Importance of Attention Matrices\", highlights the impact of fine-"},"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":"2410.02247","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-10-03T06:37:37Z","cross_cats_sorted":[],"title_canon_sha256":"be61b15fbbb15ce1d66885d3421ab5013a33cf587dd2be8e8da6a4b0630032d5","abstract_canon_sha256":"5ed16e778b3c95fdd031ffe35804575658ff1f96bbd3a07ce8aadc9ab7d9c043"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:02:39.792574Z","signature_b64":"OguDtTlLHjyBNmHYbN7nTXpj2ztZn8n8tCp6NemNwpfz8e1pES4ekCCtqq89XTWqwFgZpfy2mi/Hk7rSUx2qBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c5be7123ebe5c4164d91a02cfc8425950d94b57ab4e0694c13d9e4ab40d13c0e","last_reissued_at":"2026-07-05T11:02:39.792103Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:02:39.792103Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Theoretical Insights into Fine-Tuning Attention Mechanism: Generalization and Optimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Bin Wang, Gengze Xu, Hongjin Qian, Jian Luan, Wei Liu, Xiaolin Hu, Xinhao Yao, Yong Liu","submitted_at":"2024-10-03T06:37:37Z","abstract_excerpt":"Large Language Models (LLMs), built on Transformer architectures, exhibit remarkable generalization across a wide range of tasks. However, fine-tuning these models for specific tasks remains resource-intensive due to their extensive parameterization. In this paper, we explore two remarkable phenomena related to the attention mechanism during the fine-tuning of LLMs (where $\\mathbf{W}_q$, $\\mathbf{W}_k$, and $\\mathbf{W}_v$ denote the weights of the query, key, and value layers, respectively). The first phenomenon, termed \"Unequal Importance of Attention Matrices\", highlights the impact of fine-"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.02247","kind":"arxiv","version":3},"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/2410.02247/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":"2410.02247","created_at":"2026-07-05T11:02:39.792159+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.02247v3","created_at":"2026-07-05T11:02:39.792159+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.02247","created_at":"2026-07-05T11:02:39.792159+00:00"},{"alias_kind":"pith_short_12","alias_value":"YW7HCI7L4XCB","created_at":"2026-07-05T11:02:39.792159+00:00"},{"alias_kind":"pith_short_16","alias_value":"YW7HCI7L4XCBMTMR","created_at":"2026-07-05T11:02:39.792159+00:00"},{"alias_kind":"pith_short_8","alias_value":"YW7HCI7L","created_at":"2026-07-05T11:02:39.792159+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.12851","citing_title":"Simulate, Refocus and Ensemble: An Attention-Refocusing Scheme for Domain Generalization","ref_index":51,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YW7HCI7L4XCBMTMRUAWPZBBFSU","json":"https://pith.science/pith/YW7HCI7L4XCBMTMRUAWPZBBFSU.json","graph_json":"https://pith.science/api/pith-number/YW7HCI7L4XCBMTMRUAWPZBBFSU/graph.json","events_json":"https://pith.science/api/pith-number/YW7HCI7L4XCBMTMRUAWPZBBFSU/events.json","paper":"https://pith.science/paper/YW7HCI7L"},"agent_actions":{"view_html":"https://pith.science/pith/YW7HCI7L4XCBMTMRUAWPZBBFSU","download_json":"https://pith.science/pith/YW7HCI7L4XCBMTMRUAWPZBBFSU.json","view_paper":"https://pith.science/paper/YW7HCI7L","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.02247&json=true","fetch_graph":"https://pith.science/api/pith-number/YW7HCI7L4XCBMTMRUAWPZBBFSU/graph.json","fetch_events":"https://pith.science/api/pith-number/YW7HCI7L4XCBMTMRUAWPZBBFSU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YW7HCI7L4XCBMTMRUAWPZBBFSU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YW7HCI7L4XCBMTMRUAWPZBBFSU/action/storage_attestation","attest_author":"https://pith.science/pith/YW7HCI7L4XCBMTMRUAWPZBBFSU/action/author_attestation","sign_citation":"https://pith.science/pith/YW7HCI7L4XCBMTMRUAWPZBBFSU/action/citation_signature","submit_replication":"https://pith.science/pith/YW7HCI7L4XCBMTMRUAWPZBBFSU/action/replication_record"}},"created_at":"2026-07-05T11:02:39.792159+00:00","updated_at":"2026-07-05T11:02:39.792159+00:00"}