{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:3P2VQI5NZUJPVBVQSIPPXKR6U7","short_pith_number":"pith:3P2VQI5N","schema_version":"1.0","canonical_sha256":"dbf55823adcd12fa86b0921efbaa3ea7e7e6e626442fae486fbdb01a3ca7a69a","source":{"kind":"arxiv","id":"2110.02732","version":4},"attestation_state":"computed","paper":{"title":"On Margin Maximization in Linear and ReLU Networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Gal Vardi, Nathan Srebro, Ohad Shamir","submitted_at":"2021-10-06T13:27:23Z","abstract_excerpt":"The implicit bias of neural networks has been extensively studied in recent years. Lyu and Li [2019] showed that in homogeneous networks trained with the exponential or the logistic loss, gradient flow converges to a KKT point of the max margin problem in the parameter space. However, that leaves open the question of whether this point will generally be an actual optimum of the max margin problem. In this paper, we study this question in detail, for several neural network architectures involving linear and ReLU activations. Perhaps surprisingly, we show that in many cases, the KKT point is not"},"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":"2110.02732","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-10-06T13:27:23Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"a1e3ec8df7eeb2c0d3e45fdd5a6bf5061868a52282691624eac477e30d2fc96b","abstract_canon_sha256":"3498c8682d30c7906e54683cbbc9a735eb1f8a2084034ea4445e2327acdd2a20"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:02:23.602764Z","signature_b64":"9f79oNj93WcOIzPoZKuHvczkeNAjHgGTWpRWjVIuxQF/CnXnKb6LA9bhtF1n3kvi53IA4YdbKxiysyTIpPTxBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dbf55823adcd12fa86b0921efbaa3ea7e7e6e626442fae486fbdb01a3ca7a69a","last_reissued_at":"2026-07-05T05:02:23.602355Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:02:23.602355Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On Margin Maximization in Linear and ReLU Networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Gal Vardi, Nathan Srebro, Ohad Shamir","submitted_at":"2021-10-06T13:27:23Z","abstract_excerpt":"The implicit bias of neural networks has been extensively studied in recent years. Lyu and Li [2019] showed that in homogeneous networks trained with the exponential or the logistic loss, gradient flow converges to a KKT point of the max margin problem in the parameter space. However, that leaves open the question of whether this point will generally be an actual optimum of the max margin problem. In this paper, we study this question in detail, for several neural network architectures involving linear and ReLU activations. Perhaps surprisingly, we show that in many cases, the KKT point is not"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2110.02732","kind":"arxiv","version":4},"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/2110.02732/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":"2110.02732","created_at":"2026-07-05T05:02:23.602422+00:00"},{"alias_kind":"arxiv_version","alias_value":"2110.02732v4","created_at":"2026-07-05T05:02:23.602422+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2110.02732","created_at":"2026-07-05T05:02:23.602422+00:00"},{"alias_kind":"pith_short_12","alias_value":"3P2VQI5NZUJP","created_at":"2026-07-05T05:02:23.602422+00:00"},{"alias_kind":"pith_short_16","alias_value":"3P2VQI5NZUJPVBVQ","created_at":"2026-07-05T05:02:23.602422+00:00"},{"alias_kind":"pith_short_8","alias_value":"3P2VQI5N","created_at":"2026-07-05T05:02:23.602422+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.05436","citing_title":"Estimating Implicit Regularization in Deep Learning","ref_index":39,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3P2VQI5NZUJPVBVQSIPPXKR6U7","json":"https://pith.science/pith/3P2VQI5NZUJPVBVQSIPPXKR6U7.json","graph_json":"https://pith.science/api/pith-number/3P2VQI5NZUJPVBVQSIPPXKR6U7/graph.json","events_json":"https://pith.science/api/pith-number/3P2VQI5NZUJPVBVQSIPPXKR6U7/events.json","paper":"https://pith.science/paper/3P2VQI5N"},"agent_actions":{"view_html":"https://pith.science/pith/3P2VQI5NZUJPVBVQSIPPXKR6U7","download_json":"https://pith.science/pith/3P2VQI5NZUJPVBVQSIPPXKR6U7.json","view_paper":"https://pith.science/paper/3P2VQI5N","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2110.02732&json=true","fetch_graph":"https://pith.science/api/pith-number/3P2VQI5NZUJPVBVQSIPPXKR6U7/graph.json","fetch_events":"https://pith.science/api/pith-number/3P2VQI5NZUJPVBVQSIPPXKR6U7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3P2VQI5NZUJPVBVQSIPPXKR6U7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3P2VQI5NZUJPVBVQSIPPXKR6U7/action/storage_attestation","attest_author":"https://pith.science/pith/3P2VQI5NZUJPVBVQSIPPXKR6U7/action/author_attestation","sign_citation":"https://pith.science/pith/3P2VQI5NZUJPVBVQSIPPXKR6U7/action/citation_signature","submit_replication":"https://pith.science/pith/3P2VQI5NZUJPVBVQSIPPXKR6U7/action/replication_record"}},"created_at":"2026-07-05T05:02:23.602422+00:00","updated_at":"2026-07-05T05:02:23.602422+00:00"}