{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2018:X6C7HE6GW3MZ7Z3NDLH3MRE6VI","short_pith_number":"pith:X6C7HE6G","schema_version":"1.0","canonical_sha256":"bf85f393c6b6d99fe76d1acfb6449eaa312766b32db4a95759d6b8ee9c74d611","source":{"kind":"arxiv","id":"1808.07217","version":6},"attestation_state":"computed","paper":{"title":"Don't Use Large Mini-Batches, Use Local SGD","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Kumar Kshitij Patel, Martin Jaggi, Sebastian U. Stich, Tao Lin","submitted_at":"2018-08-22T04:50:55Z","abstract_excerpt":"Mini-batch stochastic gradient methods (SGD) are state of the art for distributed training of deep neural networks. Drastic increases in the mini-batch sizes have lead to key efficiency and scalability gains in recent years. However, progress faces a major roadblock, as models trained with large batches often do not generalize well, i.e. they do not show good accuracy on new data. As a remedy, we propose a \\emph{post-local} SGD and show that it significantly improves the generalization performance compared to large-batch training on standard benchmarks while enjoying the same efficiency (time-"},"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":"1808.07217","kind":"arxiv","version":6},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2018-08-22T04:50:55Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"9406e39837368c7f5ed49513798826363b768aeffd47ad72a4ab6af72d48d655","abstract_canon_sha256":"0129bd50d66cbe7b561b5269da5d44840bfc61ba1cee2cd8c586e8fd5d499f19"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T00:40:53.188114Z","signature_b64":"glEz0gsypUZFSgy8Qc5iftJel5y9vm10w7SupUVctGjnNQ55XsHRaa4mqxN+Tn69UgtwaXv7SzBAM8sMYEW4BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bf85f393c6b6d99fe76d1acfb6449eaa312766b32db4a95759d6b8ee9c74d611","last_reissued_at":"2026-07-05T00:40:53.187609Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T00:40:53.187609Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Don't Use Large Mini-Batches, Use Local SGD","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Kumar Kshitij Patel, Martin Jaggi, Sebastian U. Stich, Tao Lin","submitted_at":"2018-08-22T04:50:55Z","abstract_excerpt":"Mini-batch stochastic gradient methods (SGD) are state of the art for distributed training of deep neural networks. Drastic increases in the mini-batch sizes have lead to key efficiency and scalability gains in recent years. However, progress faces a major roadblock, as models trained with large batches often do not generalize well, i.e. they do not show good accuracy on new data. As a remedy, we propose a \\emph{post-local} SGD and show that it significantly improves the generalization performance compared to large-batch training on standard benchmarks while enjoying the same efficiency (time-"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1808.07217","kind":"arxiv","version":6},"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/1808.07217/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":"1808.07217","created_at":"2026-07-05T00:40:53.187672+00:00"},{"alias_kind":"arxiv_version","alias_value":"1808.07217v6","created_at":"2026-07-05T00:40:53.187672+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1808.07217","created_at":"2026-07-05T00:40:53.187672+00:00"},{"alias_kind":"pith_short_12","alias_value":"X6C7HE6GW3MZ","created_at":"2026-07-05T00:40:53.187672+00:00"},{"alias_kind":"pith_short_16","alias_value":"X6C7HE6GW3MZ7Z3N","created_at":"2026-07-05T00:40:53.187672+00:00"},{"alias_kind":"pith_short_8","alias_value":"X6C7HE6G","created_at":"2026-07-05T00:40:53.187672+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.03498","citing_title":"Demystifying Pipeline Parallelism: First Theory for PipeDream","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01128","citing_title":"Local MixVR: Breaking the Communication-Sample Dependence in Distributed Learning","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"1907.03909","citing_title":"Collaborative Machine Learning at the Wireless Edge with Blind Transmitters","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2003.00295","citing_title":"Adaptive Federated Optimization","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10205","citing_title":"Unveiling High-Probability Generalization in Decentralized SGD","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01701","citing_title":"Stability and Generalization for Decentralized Markov SGD","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12768","citing_title":"Rethinking the Personalized Relaxed Initialization in the Federated Learning: Consistency and Generalization","ref_index":8,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/X6C7HE6GW3MZ7Z3NDLH3MRE6VI","json":"https://pith.science/pith/X6C7HE6GW3MZ7Z3NDLH3MRE6VI.json","graph_json":"https://pith.science/api/pith-number/X6C7HE6GW3MZ7Z3NDLH3MRE6VI/graph.json","events_json":"https://pith.science/api/pith-number/X6C7HE6GW3MZ7Z3NDLH3MRE6VI/events.json","paper":"https://pith.science/paper/X6C7HE6G"},"agent_actions":{"view_html":"https://pith.science/pith/X6C7HE6GW3MZ7Z3NDLH3MRE6VI","download_json":"https://pith.science/pith/X6C7HE6GW3MZ7Z3NDLH3MRE6VI.json","view_paper":"https://pith.science/paper/X6C7HE6G","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1808.07217&json=true","fetch_graph":"https://pith.science/api/pith-number/X6C7HE6GW3MZ7Z3NDLH3MRE6VI/graph.json","fetch_events":"https://pith.science/api/pith-number/X6C7HE6GW3MZ7Z3NDLH3MRE6VI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/X6C7HE6GW3MZ7Z3NDLH3MRE6VI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/X6C7HE6GW3MZ7Z3NDLH3MRE6VI/action/storage_attestation","attest_author":"https://pith.science/pith/X6C7HE6GW3MZ7Z3NDLH3MRE6VI/action/author_attestation","sign_citation":"https://pith.science/pith/X6C7HE6GW3MZ7Z3NDLH3MRE6VI/action/citation_signature","submit_replication":"https://pith.science/pith/X6C7HE6GW3MZ7Z3NDLH3MRE6VI/action/replication_record"}},"created_at":"2026-07-05T00:40:53.187672+00:00","updated_at":"2026-07-05T00:40:53.187672+00:00"}