{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:GY5VBGTQ37N6F7FZ4AAJACMNMI","short_pith_number":"pith:GY5VBGTQ","schema_version":"1.0","canonical_sha256":"363b509a70dfdbe2fcb9e00090098d6213f10f7668d5eea762ad4c8824d8936b","source":{"kind":"arxiv","id":"2403.08699","version":1},"attestation_state":"computed","paper":{"title":"Implicit Regularization of Gradient Flow on One-Layer Softmax Attention","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Harrison H. Zhou, Heejune Sheen, Siyu Chen, Tianhao Wang","submitted_at":"2024-03-13T17:02:27Z","abstract_excerpt":"We study gradient flow on the exponential loss for a classification problem with a one-layer softmax attention model, where the key and query weight matrices are trained separately. Under a separability assumption on the data, we show that when gradient flow achieves the minimal loss value, it further implicitly minimizes the nuclear norm of the product of the key and query weight matrices. Such implicit regularization can be described by a Support Vector Machine (SVM) problem with respect to the attention weights. This finding contrasts with prior results showing that the gradient descent ind"},"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":"2403.08699","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-03-13T17:02:27Z","cross_cats_sorted":["cs.AI","math.OC","stat.ML"],"title_canon_sha256":"820748c4db5d62466873ba92e9dd334b64b6164795a97019d0115cfba0091da9","abstract_canon_sha256":"4f79993c254686a6268a32e20679b5873753c53503e39972db47adc0a3d98406"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:55:42.589494Z","signature_b64":"oiiMDlWvsYHtmb1ssDUhAzPKzxyv0uAuBOz3gP9EaAxxtv0/m+ZFwHc0FkYveQuVqHpJf6YH1Ul1Kr5ukwNfCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"363b509a70dfdbe2fcb9e00090098d6213f10f7668d5eea762ad4c8824d8936b","last_reissued_at":"2026-07-05T07:55:42.588812Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:55:42.588812Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Implicit Regularization of Gradient Flow on One-Layer Softmax Attention","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Harrison H. Zhou, Heejune Sheen, Siyu Chen, Tianhao Wang","submitted_at":"2024-03-13T17:02:27Z","abstract_excerpt":"We study gradient flow on the exponential loss for a classification problem with a one-layer softmax attention model, where the key and query weight matrices are trained separately. Under a separability assumption on the data, we show that when gradient flow achieves the minimal loss value, it further implicitly minimizes the nuclear norm of the product of the key and query weight matrices. Such implicit regularization can be described by a Support Vector Machine (SVM) problem with respect to the attention weights. This finding contrasts with prior results showing that the gradient descent ind"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.08699","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/2403.08699/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":"2403.08699","created_at":"2026-07-05T07:55:42.588892+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.08699v1","created_at":"2026-07-05T07:55:42.588892+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.08699","created_at":"2026-07-05T07:55:42.588892+00:00"},{"alias_kind":"pith_short_12","alias_value":"GY5VBGTQ37N6","created_at":"2026-07-05T07:55:42.588892+00:00"},{"alias_kind":"pith_short_16","alias_value":"GY5VBGTQ37N6F7FZ","created_at":"2026-07-05T07:55:42.588892+00:00"},{"alias_kind":"pith_short_8","alias_value":"GY5VBGTQ","created_at":"2026-07-05T07:55:42.588892+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.12058","citing_title":"Phase Transitions in Attention: A Bayesian Theory of Copy Head Emergence","ref_index":61,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21217","citing_title":"Federated LoRA Fine-Tuning for LLMs via Collaborative Alignment","ref_index":27,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GY5VBGTQ37N6F7FZ4AAJACMNMI","json":"https://pith.science/pith/GY5VBGTQ37N6F7FZ4AAJACMNMI.json","graph_json":"https://pith.science/api/pith-number/GY5VBGTQ37N6F7FZ4AAJACMNMI/graph.json","events_json":"https://pith.science/api/pith-number/GY5VBGTQ37N6F7FZ4AAJACMNMI/events.json","paper":"https://pith.science/paper/GY5VBGTQ"},"agent_actions":{"view_html":"https://pith.science/pith/GY5VBGTQ37N6F7FZ4AAJACMNMI","download_json":"https://pith.science/pith/GY5VBGTQ37N6F7FZ4AAJACMNMI.json","view_paper":"https://pith.science/paper/GY5VBGTQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.08699&json=true","fetch_graph":"https://pith.science/api/pith-number/GY5VBGTQ37N6F7FZ4AAJACMNMI/graph.json","fetch_events":"https://pith.science/api/pith-number/GY5VBGTQ37N6F7FZ4AAJACMNMI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GY5VBGTQ37N6F7FZ4AAJACMNMI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GY5VBGTQ37N6F7FZ4AAJACMNMI/action/storage_attestation","attest_author":"https://pith.science/pith/GY5VBGTQ37N6F7FZ4AAJACMNMI/action/author_attestation","sign_citation":"https://pith.science/pith/GY5VBGTQ37N6F7FZ4AAJACMNMI/action/citation_signature","submit_replication":"https://pith.science/pith/GY5VBGTQ37N6F7FZ4AAJACMNMI/action/replication_record"}},"created_at":"2026-07-05T07:55:42.588892+00:00","updated_at":"2026-07-05T07:55:42.588892+00:00"}