{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:XMNVQ56B2PE3EJASNDHNK2NHHN","short_pith_number":"pith:XMNVQ56B","schema_version":"1.0","canonical_sha256":"bb1b5877c1d3c9b2241268ced569a73b5ce9216ba50788f9822988ea41054df7","source":{"kind":"arxiv","id":"2301.02312","version":2},"attestation_state":"computed","paper":{"title":"Training trajectories, mini-batch losses and the curious role of the learning rate","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Andrey Zhmoginov, Mark Sandler, Max Vladymyrov, Nolan Miller","submitted_at":"2023-01-05T21:58:46Z","abstract_excerpt":"Stochastic gradient descent plays a fundamental role in nearly all applications of deep learning. However its ability to converge to a global minimum remains shrouded in mystery. In this paper we propose to study the behavior of the loss function on fixed mini-batches along SGD trajectories. We show that the loss function on a fixed batch appears to be remarkably convex-like. In particular for ResNet the loss for any fixed mini-batch can be accurately modeled by a quadratic function and a very low loss value can be reached in just one step of gradient descent with sufficiently large learning r"},"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":"2301.02312","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-01-05T21:58:46Z","cross_cats_sorted":[],"title_canon_sha256":"99b478b45bf7dda793a997e72487cb035cab717bde3c3e21b285fd3e816c10c1","abstract_canon_sha256":"0799f3823022ff304dff35d3e7dd826687ea3b6b198a7f01fbf6f05211a18c13"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:37:56.815109Z","signature_b64":"WyVqjh0W1dDh4qmJXQGlcsEdpE9pt25MkgMLeanVkiETFML8wQFlSdPDM1HmZhRpltG4IGf5foGF8yDaTXJ1Bg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bb1b5877c1d3c9b2241268ced569a73b5ce9216ba50788f9822988ea41054df7","last_reissued_at":"2026-07-05T05:37:56.814582Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:37:56.814582Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Training trajectories, mini-batch losses and the curious role of the learning rate","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Andrey Zhmoginov, Mark Sandler, Max Vladymyrov, Nolan Miller","submitted_at":"2023-01-05T21:58:46Z","abstract_excerpt":"Stochastic gradient descent plays a fundamental role in nearly all applications of deep learning. However its ability to converge to a global minimum remains shrouded in mystery. In this paper we propose to study the behavior of the loss function on fixed mini-batches along SGD trajectories. We show that the loss function on a fixed batch appears to be remarkably convex-like. In particular for ResNet the loss for any fixed mini-batch can be accurately modeled by a quadratic function and a very low loss value can be reached in just one step of gradient descent with sufficiently large learning r"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2301.02312","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/2301.02312/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":"2301.02312","created_at":"2026-07-05T05:37:56.814642+00:00"},{"alias_kind":"arxiv_version","alias_value":"2301.02312v2","created_at":"2026-07-05T05:37:56.814642+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2301.02312","created_at":"2026-07-05T05:37:56.814642+00:00"},{"alias_kind":"pith_short_12","alias_value":"XMNVQ56B2PE3","created_at":"2026-07-05T05:37:56.814642+00:00"},{"alias_kind":"pith_short_16","alias_value":"XMNVQ56B2PE3EJAS","created_at":"2026-07-05T05:37:56.814642+00:00"},{"alias_kind":"pith_short_8","alias_value":"XMNVQ56B","created_at":"2026-07-05T05:37:56.814642+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.19095","citing_title":"ScheduleFree+: Scaling Learning-Rate-Free & Schedule-Free Learning to Large Language Models","ref_index":29,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XMNVQ56B2PE3EJASNDHNK2NHHN","json":"https://pith.science/pith/XMNVQ56B2PE3EJASNDHNK2NHHN.json","graph_json":"https://pith.science/api/pith-number/XMNVQ56B2PE3EJASNDHNK2NHHN/graph.json","events_json":"https://pith.science/api/pith-number/XMNVQ56B2PE3EJASNDHNK2NHHN/events.json","paper":"https://pith.science/paper/XMNVQ56B"},"agent_actions":{"view_html":"https://pith.science/pith/XMNVQ56B2PE3EJASNDHNK2NHHN","download_json":"https://pith.science/pith/XMNVQ56B2PE3EJASNDHNK2NHHN.json","view_paper":"https://pith.science/paper/XMNVQ56B","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2301.02312&json=true","fetch_graph":"https://pith.science/api/pith-number/XMNVQ56B2PE3EJASNDHNK2NHHN/graph.json","fetch_events":"https://pith.science/api/pith-number/XMNVQ56B2PE3EJASNDHNK2NHHN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XMNVQ56B2PE3EJASNDHNK2NHHN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XMNVQ56B2PE3EJASNDHNK2NHHN/action/storage_attestation","attest_author":"https://pith.science/pith/XMNVQ56B2PE3EJASNDHNK2NHHN/action/author_attestation","sign_citation":"https://pith.science/pith/XMNVQ56B2PE3EJASNDHNK2NHHN/action/citation_signature","submit_replication":"https://pith.science/pith/XMNVQ56B2PE3EJASNDHNK2NHHN/action/replication_record"}},"created_at":"2026-07-05T05:37:56.814642+00:00","updated_at":"2026-07-05T05:37:56.814642+00:00"}