{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:ABJSXHJTR2HOV7QRGF6DO5MHUP","short_pith_number":"pith:ABJSXHJT","schema_version":"1.0","canonical_sha256":"00532b9d338e8eeafe11317c377587a3f06e07c4b34ab1d9f7b89f40c2e144e6","source":{"kind":"arxiv","id":"2206.07085","version":3},"attestation_state":"computed","paper":{"title":"Understanding the Generalization Benefit of Normalization Layers: Sharpness Reduction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Kaifeng Lyu, Sanjeev Arora, Zhiyuan Li","submitted_at":"2022-06-14T18:19:05Z","abstract_excerpt":"Normalization layers (e.g., Batch Normalization, Layer Normalization) were introduced to help with optimization difficulties in very deep nets, but they clearly also help generalization, even in not-so-deep nets. Motivated by the long-held belief that flatter minima lead to better generalization, this paper gives mathematical analysis and supporting experiments suggesting that normalization (together with accompanying weight-decay) encourages GD to reduce the sharpness of loss surface. Here \"sharpness\" is carefully defined given that the loss is scale-invariant, a known consequence of normaliz"},"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":"2206.07085","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2022-06-14T18:19:05Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"eaa91422cfc3826ef7790843a6b555cdade771f048cf1642378fb04f795c796a","abstract_canon_sha256":"4d566c012e6bba23fc71c11c3a136b86186a714ff3c722b4474d4518af31fadc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:33:12.754016Z","signature_b64":"PSYShar1YTYQyNkep2i2IEoFNtTB0knRJOkA1lhzxC7wOcthQlwd722v1XxRthR7PBi5cLFo7//fP5Okl3RkDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"00532b9d338e8eeafe11317c377587a3f06e07c4b34ab1d9f7b89f40c2e144e6","last_reissued_at":"2026-07-05T05:33:12.753512Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:33:12.753512Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding the Generalization Benefit of Normalization Layers: Sharpness Reduction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Kaifeng Lyu, Sanjeev Arora, Zhiyuan Li","submitted_at":"2022-06-14T18:19:05Z","abstract_excerpt":"Normalization layers (e.g., Batch Normalization, Layer Normalization) were introduced to help with optimization difficulties in very deep nets, but they clearly also help generalization, even in not-so-deep nets. Motivated by the long-held belief that flatter minima lead to better generalization, this paper gives mathematical analysis and supporting experiments suggesting that normalization (together with accompanying weight-decay) encourages GD to reduce the sharpness of loss surface. Here \"sharpness\" is carefully defined given that the loss is scale-invariant, a known consequence of normaliz"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2206.07085","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/2206.07085/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":"2206.07085","created_at":"2026-07-05T05:33:12.753572+00:00"},{"alias_kind":"arxiv_version","alias_value":"2206.07085v3","created_at":"2026-07-05T05:33:12.753572+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2206.07085","created_at":"2026-07-05T05:33:12.753572+00:00"},{"alias_kind":"pith_short_12","alias_value":"ABJSXHJTR2HO","created_at":"2026-07-05T05:33:12.753572+00:00"},{"alias_kind":"pith_short_16","alias_value":"ABJSXHJTR2HOV7QR","created_at":"2026-07-05T05:33:12.753572+00:00"},{"alias_kind":"pith_short_8","alias_value":"ABJSXHJT","created_at":"2026-07-05T05:33:12.753572+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.28662","citing_title":"Closed-Form Steepest Descent Direction toward Flat Minima: Reducing Upper Bounds on the Loss Hessian Eigenspectrum in Neural Networks","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10202","citing_title":"Wolkowicz-Styan Upper Bound on the Hessian Eigenspectrum for Cross-Entropy Loss in Nonlinear Smooth Neural Networks","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ABJSXHJTR2HOV7QRGF6DO5MHUP","json":"https://pith.science/pith/ABJSXHJTR2HOV7QRGF6DO5MHUP.json","graph_json":"https://pith.science/api/pith-number/ABJSXHJTR2HOV7QRGF6DO5MHUP/graph.json","events_json":"https://pith.science/api/pith-number/ABJSXHJTR2HOV7QRGF6DO5MHUP/events.json","paper":"https://pith.science/paper/ABJSXHJT"},"agent_actions":{"view_html":"https://pith.science/pith/ABJSXHJTR2HOV7QRGF6DO5MHUP","download_json":"https://pith.science/pith/ABJSXHJTR2HOV7QRGF6DO5MHUP.json","view_paper":"https://pith.science/paper/ABJSXHJT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2206.07085&json=true","fetch_graph":"https://pith.science/api/pith-number/ABJSXHJTR2HOV7QRGF6DO5MHUP/graph.json","fetch_events":"https://pith.science/api/pith-number/ABJSXHJTR2HOV7QRGF6DO5MHUP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ABJSXHJTR2HOV7QRGF6DO5MHUP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ABJSXHJTR2HOV7QRGF6DO5MHUP/action/storage_attestation","attest_author":"https://pith.science/pith/ABJSXHJTR2HOV7QRGF6DO5MHUP/action/author_attestation","sign_citation":"https://pith.science/pith/ABJSXHJTR2HOV7QRGF6DO5MHUP/action/citation_signature","submit_replication":"https://pith.science/pith/ABJSXHJTR2HOV7QRGF6DO5MHUP/action/replication_record"}},"created_at":"2026-07-05T05:33:12.753572+00:00","updated_at":"2026-07-05T05:33:12.753572+00:00"}