{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:EJYR2UKB6ZNB3MGKU7LJHA5OFK","short_pith_number":"pith:EJYR2UKB","schema_version":"1.0","canonical_sha256":"22711d5141f65a1db0caa7d69383ae2aa16d9b5c6de63146aeaa2b86c68ef666","source":{"kind":"arxiv","id":"2211.09359","version":2},"attestation_state":"computed","paper":{"title":"How to Fine-Tune Vision Models with SGD","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Ananya Kumar, Ruoqi Shen, Sebastien Bubeck, Suriya Gunasekar","submitted_at":"2022-11-17T06:18:45Z","abstract_excerpt":"SGD and AdamW are the two most used optimizers for fine-tuning large neural networks in computer vision. When the two methods perform the same, SGD is preferable because it uses less memory (12 bytes/parameter with momentum and 8 bytes/parameter without) than AdamW (16 bytes/parameter). However, on a suite of downstream tasks, especially those with distribution shifts, we find that fine-tuning with AdamW performs substantially better than SGD on modern Vision Transformer and ConvNeXt models. We find that large gaps in performance between SGD and AdamW occur when the fine-tuning gradients in th"},"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":"2211.09359","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2022-11-17T06:18:45Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"51c13201460228afa46232c6f68eb10e0bc03a635cc4e94663b32f250aac8168","abstract_canon_sha256":"04aa6b98a3aa7b83021382b840b5ea90e93c354368366418a08f41ee5180d51f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:58:54.064978Z","signature_b64":"EpM2dpQrEdzJtzAIuwezZxMCjflfM2RdfD/sXlzQy5M5wjdt76Wd1F/VP5HyDmNFGpfZ4Oi9mMCtAb8sqTmTBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"22711d5141f65a1db0caa7d69383ae2aa16d9b5c6de63146aeaa2b86c68ef666","last_reissued_at":"2026-07-05T06:58:54.064452Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:58:54.064452Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How to Fine-Tune Vision Models with SGD","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Ananya Kumar, Ruoqi Shen, Sebastien Bubeck, Suriya Gunasekar","submitted_at":"2022-11-17T06:18:45Z","abstract_excerpt":"SGD and AdamW are the two most used optimizers for fine-tuning large neural networks in computer vision. When the two methods perform the same, SGD is preferable because it uses less memory (12 bytes/parameter with momentum and 8 bytes/parameter without) than AdamW (16 bytes/parameter). However, on a suite of downstream tasks, especially those with distribution shifts, we find that fine-tuning with AdamW performs substantially better than SGD on modern Vision Transformer and ConvNeXt models. We find that large gaps in performance between SGD and AdamW occur when the fine-tuning gradients in th"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2211.09359","kind":"arxiv","version":2},"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/2211.09359/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":"2211.09359","created_at":"2026-07-05T06:58:54.064512+00:00"},{"alias_kind":"arxiv_version","alias_value":"2211.09359v2","created_at":"2026-07-05T06:58:54.064512+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2211.09359","created_at":"2026-07-05T06:58:54.064512+00:00"},{"alias_kind":"pith_short_12","alias_value":"EJYR2UKB6ZNB","created_at":"2026-07-05T06:58:54.064512+00:00"},{"alias_kind":"pith_short_16","alias_value":"EJYR2UKB6ZNB3MGK","created_at":"2026-07-05T06:58:54.064512+00:00"},{"alias_kind":"pith_short_8","alias_value":"EJYR2UKB","created_at":"2026-07-05T06:58:54.064512+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2506.12543","citing_title":"Is your batch size the problem? Revisiting the Adam-SGD gap in language modeling","ref_index":2014,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EJYR2UKB6ZNB3MGKU7LJHA5OFK","json":"https://pith.science/pith/EJYR2UKB6ZNB3MGKU7LJHA5OFK.json","graph_json":"https://pith.science/api/pith-number/EJYR2UKB6ZNB3MGKU7LJHA5OFK/graph.json","events_json":"https://pith.science/api/pith-number/EJYR2UKB6ZNB3MGKU7LJHA5OFK/events.json","paper":"https://pith.science/paper/EJYR2UKB"},"agent_actions":{"view_html":"https://pith.science/pith/EJYR2UKB6ZNB3MGKU7LJHA5OFK","download_json":"https://pith.science/pith/EJYR2UKB6ZNB3MGKU7LJHA5OFK.json","view_paper":"https://pith.science/paper/EJYR2UKB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2211.09359&json=true","fetch_graph":"https://pith.science/api/pith-number/EJYR2UKB6ZNB3MGKU7LJHA5OFK/graph.json","fetch_events":"https://pith.science/api/pith-number/EJYR2UKB6ZNB3MGKU7LJHA5OFK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EJYR2UKB6ZNB3MGKU7LJHA5OFK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EJYR2UKB6ZNB3MGKU7LJHA5OFK/action/storage_attestation","attest_author":"https://pith.science/pith/EJYR2UKB6ZNB3MGKU7LJHA5OFK/action/author_attestation","sign_citation":"https://pith.science/pith/EJYR2UKB6ZNB3MGKU7LJHA5OFK/action/citation_signature","submit_replication":"https://pith.science/pith/EJYR2UKB6ZNB3MGKU7LJHA5OFK/action/replication_record"}},"created_at":"2026-07-05T06:58:54.064512+00:00","updated_at":"2026-07-05T06:58:54.064512+00:00"}