{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:YUZYU4ZI3MUGWXT5SN55ZZL5PJ","short_pith_number":"pith:YUZYU4ZI","schema_version":"1.0","canonical_sha256":"c5338a7328db286b5e7d937bdce57d7a49673fbf682da33ebafc8625efc0722c","source":{"kind":"arxiv","id":"2206.07638","version":2},"attestation_state":"computed","paper":{"title":"Asynchronous SGD Beats Minibatch SGD Under Arbitrary Delays","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.DC","cs.LG"],"primary_cat":"math.OC","authors_text":"Blake Woodworth, Francis Bach, Konstantin Mishchenko, Mathieu Even","submitted_at":"2022-06-15T16:28:37Z","abstract_excerpt":"The existing analysis of asynchronous stochastic gradient descent (SGD) degrades dramatically when any delay is large, giving the impression that performance depends primarily on the delay. On the contrary, we prove much better guarantees for the same asynchronous SGD algorithm regardless of the delays in the gradients, depending instead just on the number of parallel devices used to implement the algorithm. Our guarantees are strictly better than the existing analyses, and we also argue that asynchronous SGD outperforms synchronous minibatch SGD in the settings we consider. For our analysis, "},"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":"2206.07638","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"math.OC","submitted_at":"2022-06-15T16:28:37Z","cross_cats_sorted":["cs.DC","cs.LG"],"title_canon_sha256":"13b372ddea4de898fa1474ab42aa0f16c37359ae46bb6c38ff5641b1fd951186","abstract_canon_sha256":"c579c7e10da0eb7e64ba538615b06da6cd9f6839dbfaf490cab795f71b6960cb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:02:46.328606Z","signature_b64":"GptSgtg7vrY7Rep9WasNG4c6ENVkIRcOs0B2UPR83htFk19w1pOoRKsgc1RKge9VFGwdWlLYBzivTw5BbxiNDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c5338a7328db286b5e7d937bdce57d7a49673fbf682da33ebafc8625efc0722c","last_reissued_at":"2026-07-05T06:02:46.328116Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:02:46.328116Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Asynchronous SGD Beats Minibatch SGD Under Arbitrary Delays","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.DC","cs.LG"],"primary_cat":"math.OC","authors_text":"Blake Woodworth, Francis Bach, Konstantin Mishchenko, Mathieu Even","submitted_at":"2022-06-15T16:28:37Z","abstract_excerpt":"The existing analysis of asynchronous stochastic gradient descent (SGD) degrades dramatically when any delay is large, giving the impression that performance depends primarily on the delay. On the contrary, we prove much better guarantees for the same asynchronous SGD algorithm regardless of the delays in the gradients, depending instead just on the number of parallel devices used to implement the algorithm. Our guarantees are strictly better than the existing analyses, and we also argue that asynchronous SGD outperforms synchronous minibatch SGD in the settings we consider. For our analysis, "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2206.07638","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/2206.07638/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":"2206.07638","created_at":"2026-07-05T06:02:46.328175+00:00"},{"alias_kind":"arxiv_version","alias_value":"2206.07638v2","created_at":"2026-07-05T06:02:46.328175+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2206.07638","created_at":"2026-07-05T06:02:46.328175+00:00"},{"alias_kind":"pith_short_12","alias_value":"YUZYU4ZI3MUG","created_at":"2026-07-05T06:02:46.328175+00:00"},{"alias_kind":"pith_short_16","alias_value":"YUZYU4ZI3MUGWXT5","created_at":"2026-07-05T06:02:46.328175+00:00"},{"alias_kind":"pith_short_8","alias_value":"YUZYU4ZI","created_at":"2026-07-05T06:02:46.328175+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.09126","citing_title":"Cosine-Gated Adam-Decay: Drop-In Staleness-Aware Outer Optimization for Decoupled DiLoCo","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YUZYU4ZI3MUGWXT5SN55ZZL5PJ","json":"https://pith.science/pith/YUZYU4ZI3MUGWXT5SN55ZZL5PJ.json","graph_json":"https://pith.science/api/pith-number/YUZYU4ZI3MUGWXT5SN55ZZL5PJ/graph.json","events_json":"https://pith.science/api/pith-number/YUZYU4ZI3MUGWXT5SN55ZZL5PJ/events.json","paper":"https://pith.science/paper/YUZYU4ZI"},"agent_actions":{"view_html":"https://pith.science/pith/YUZYU4ZI3MUGWXT5SN55ZZL5PJ","download_json":"https://pith.science/pith/YUZYU4ZI3MUGWXT5SN55ZZL5PJ.json","view_paper":"https://pith.science/paper/YUZYU4ZI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2206.07638&json=true","fetch_graph":"https://pith.science/api/pith-number/YUZYU4ZI3MUGWXT5SN55ZZL5PJ/graph.json","fetch_events":"https://pith.science/api/pith-number/YUZYU4ZI3MUGWXT5SN55ZZL5PJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YUZYU4ZI3MUGWXT5SN55ZZL5PJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YUZYU4ZI3MUGWXT5SN55ZZL5PJ/action/storage_attestation","attest_author":"https://pith.science/pith/YUZYU4ZI3MUGWXT5SN55ZZL5PJ/action/author_attestation","sign_citation":"https://pith.science/pith/YUZYU4ZI3MUGWXT5SN55ZZL5PJ/action/citation_signature","submit_replication":"https://pith.science/pith/YUZYU4ZI3MUGWXT5SN55ZZL5PJ/action/replication_record"}},"created_at":"2026-07-05T06:02:46.328175+00:00","updated_at":"2026-07-05T06:02:46.328175+00:00"}