{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:AQH7M24EGQRJRUTIQLF5XKPGWX","short_pith_number":"pith:AQH7M24E","schema_version":"1.0","canonical_sha256":"040ff66b84342298d26882cbdba9e6b5e8451d38da196fb27266b8d98cb6537e","source":{"kind":"arxiv","id":"2009.11162","version":3},"attestation_state":"computed","paper":{"title":"Implicit Gradient Regularization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Benoit Dherin, David G.T. Barrett","submitted_at":"2020-09-23T14:17:53Z","abstract_excerpt":"Gradient descent can be surprisingly good at optimizing deep neural networks without overfitting and without explicit regularization. We find that the discrete steps of gradient descent implicitly regularize models by penalizing gradient descent trajectories that have large loss gradients. We call this Implicit Gradient Regularization (IGR) and we use backward error analysis to calculate the size of this regularization. We confirm empirically that implicit gradient regularization biases gradient descent toward flat minima, where test errors are small and solutions are robust to noisy parameter"},"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":"2009.11162","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2020-09-23T14:17:53Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"269ffc95c23c902ccb99c58a50209729bf96471476f090818139453608b2cfd6","abstract_canon_sha256":"22fa6bbbd310e795a1540cabca779f163608de38ce438fab6de3071e2a285260"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:41:09.062338Z","signature_b64":"E0KLf91mL9CZniLo/XoMcTC0Thz/8RWQkBxmvqXP6xGjIQrDAWCLO4+WY4jJ+oan7duSUd/HNKj9BQRgmFmYAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"040ff66b84342298d26882cbdba9e6b5e8451d38da196fb27266b8d98cb6537e","last_reissued_at":"2026-07-05T04:41:09.061866Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:41:09.061866Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Implicit Gradient Regularization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Benoit Dherin, David G.T. Barrett","submitted_at":"2020-09-23T14:17:53Z","abstract_excerpt":"Gradient descent can be surprisingly good at optimizing deep neural networks without overfitting and without explicit regularization. We find that the discrete steps of gradient descent implicitly regularize models by penalizing gradient descent trajectories that have large loss gradients. We call this Implicit Gradient Regularization (IGR) and we use backward error analysis to calculate the size of this regularization. We confirm empirically that implicit gradient regularization biases gradient descent toward flat minima, where test errors are small and solutions are robust to noisy parameter"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2009.11162","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/2009.11162/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":"2009.11162","created_at":"2026-07-05T04:41:09.061935+00:00"},{"alias_kind":"arxiv_version","alias_value":"2009.11162v3","created_at":"2026-07-05T04:41:09.061935+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2009.11162","created_at":"2026-07-05T04:41:09.061935+00:00"},{"alias_kind":"pith_short_12","alias_value":"AQH7M24EGQRJ","created_at":"2026-07-05T04:41:09.061935+00:00"},{"alias_kind":"pith_short_16","alias_value":"AQH7M24EGQRJRUTI","created_at":"2026-07-05T04:41:09.061935+00:00"},{"alias_kind":"pith_short_8","alias_value":"AQH7M24E","created_at":"2026-07-05T04:41:09.061935+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":8,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07538","citing_title":"Avoiding unsafe sets when training with Langevin Dynamics","ref_index":26,"is_internal_anchor":true},{"citing_arxiv_id":"2606.08316","citing_title":"Some Inverse Problems in Particle Physics","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07495","citing_title":"Second-Order Path Kernel Interpolation Formulas in Machine Learning","ref_index":152,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02385","citing_title":"How Optimality Structures Sparse Dictionaries: A Theory for Understanding SAE Representations","ref_index":127,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21933","citing_title":"Thermodynamic Irreversibility of Training Algorithms","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05436","citing_title":"Estimating Implicit Regularization in Deep Learning","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07886","citing_title":"Characterizing and Correcting Effective Target Shift in Online Learning","ref_index":56,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16075","citing_title":"Towards Universal Convergence of Backward Error in Linear System Solvers","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AQH7M24EGQRJRUTIQLF5XKPGWX","json":"https://pith.science/pith/AQH7M24EGQRJRUTIQLF5XKPGWX.json","graph_json":"https://pith.science/api/pith-number/AQH7M24EGQRJRUTIQLF5XKPGWX/graph.json","events_json":"https://pith.science/api/pith-number/AQH7M24EGQRJRUTIQLF5XKPGWX/events.json","paper":"https://pith.science/paper/AQH7M24E"},"agent_actions":{"view_html":"https://pith.science/pith/AQH7M24EGQRJRUTIQLF5XKPGWX","download_json":"https://pith.science/pith/AQH7M24EGQRJRUTIQLF5XKPGWX.json","view_paper":"https://pith.science/paper/AQH7M24E","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2009.11162&json=true","fetch_graph":"https://pith.science/api/pith-number/AQH7M24EGQRJRUTIQLF5XKPGWX/graph.json","fetch_events":"https://pith.science/api/pith-number/AQH7M24EGQRJRUTIQLF5XKPGWX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AQH7M24EGQRJRUTIQLF5XKPGWX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AQH7M24EGQRJRUTIQLF5XKPGWX/action/storage_attestation","attest_author":"https://pith.science/pith/AQH7M24EGQRJRUTIQLF5XKPGWX/action/author_attestation","sign_citation":"https://pith.science/pith/AQH7M24EGQRJRUTIQLF5XKPGWX/action/citation_signature","submit_replication":"https://pith.science/pith/AQH7M24EGQRJRUTIQLF5XKPGWX/action/replication_record"}},"created_at":"2026-07-05T04:41:09.061935+00:00","updated_at":"2026-07-05T04:41:09.061935+00:00"}