{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:NI3RNNCMAWL2V7C4LVDKPGZJUP","short_pith_number":"pith:NI3RNNCM","schema_version":"1.0","canonical_sha256":"6a3716b44c0597aafc5c5d46a79b29a3fab199b073feaea21851b21717bc7cc9","source":{"kind":"arxiv","id":"2107.11774","version":4},"attestation_state":"computed","paper":{"title":"SGD with a Constant Large Learning Rate Can Converge to Local Maxima","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Botao Li, James B. Simon, Liu Ziyin, Masahito Ueda","submitted_at":"2021-07-25T10:12:18Z","abstract_excerpt":"Previous works on stochastic gradient descent (SGD) often focus on its success. In this work, we construct worst-case optimization problems illustrating that, when not in the regimes that the previous works often assume, SGD can exhibit many strange and potentially undesirable behaviors. Specifically, we construct landscapes and data distributions such that (1) SGD converges to local maxima, (2) SGD escapes saddle points arbitrarily slowly, (3) SGD prefers sharp minima over flat ones, and (4) AMSGrad converges to local maxima. We also realize results in a minimal neural network-like example. O"},"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":"2107.11774","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-07-25T10:12:18Z","cross_cats_sorted":["math.OC","stat.ML"],"title_canon_sha256":"c782771213089cdc67c9b0813eb6a4b409cebd38e55f4decea8499381c33fb75","abstract_canon_sha256":"9bb80ddb4a241eece55d93c200cf05c4cbfb9eff529cba35ffff975788b63ae9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:14:33.019052Z","signature_b64":"JnS78eG+4SZBz+MOdnJFrz6t8Wi6VmOoyZKPKF/wP2DXGCbGw2NhVZeoiLy0ls1zJB1Rp1wOVyKSm1WNy0+iDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6a3716b44c0597aafc5c5d46a79b29a3fab199b073feaea21851b21717bc7cc9","last_reissued_at":"2026-07-05T06:14:33.018615Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:14:33.018615Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SGD with a Constant Large Learning Rate Can Converge to Local Maxima","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Botao Li, James B. Simon, Liu Ziyin, Masahito Ueda","submitted_at":"2021-07-25T10:12:18Z","abstract_excerpt":"Previous works on stochastic gradient descent (SGD) often focus on its success. In this work, we construct worst-case optimization problems illustrating that, when not in the regimes that the previous works often assume, SGD can exhibit many strange and potentially undesirable behaviors. Specifically, we construct landscapes and data distributions such that (1) SGD converges to local maxima, (2) SGD escapes saddle points arbitrarily slowly, (3) SGD prefers sharp minima over flat ones, and (4) AMSGrad converges to local maxima. We also realize results in a minimal neural network-like example. O"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2107.11774","kind":"arxiv","version":4},"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/2107.11774/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":"2107.11774","created_at":"2026-07-05T06:14:33.018675+00:00"},{"alias_kind":"arxiv_version","alias_value":"2107.11774v4","created_at":"2026-07-05T06:14:33.018675+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2107.11774","created_at":"2026-07-05T06:14:33.018675+00:00"},{"alias_kind":"pith_short_12","alias_value":"NI3RNNCMAWL2","created_at":"2026-07-05T06:14:33.018675+00:00"},{"alias_kind":"pith_short_16","alias_value":"NI3RNNCMAWL2V7C4","created_at":"2026-07-05T06:14:33.018675+00:00"},{"alias_kind":"pith_short_8","alias_value":"NI3RNNCM","created_at":"2026-07-05T06:14:33.018675+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NI3RNNCMAWL2V7C4LVDKPGZJUP","json":"https://pith.science/pith/NI3RNNCMAWL2V7C4LVDKPGZJUP.json","graph_json":"https://pith.science/api/pith-number/NI3RNNCMAWL2V7C4LVDKPGZJUP/graph.json","events_json":"https://pith.science/api/pith-number/NI3RNNCMAWL2V7C4LVDKPGZJUP/events.json","paper":"https://pith.science/paper/NI3RNNCM"},"agent_actions":{"view_html":"https://pith.science/pith/NI3RNNCMAWL2V7C4LVDKPGZJUP","download_json":"https://pith.science/pith/NI3RNNCMAWL2V7C4LVDKPGZJUP.json","view_paper":"https://pith.science/paper/NI3RNNCM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2107.11774&json=true","fetch_graph":"https://pith.science/api/pith-number/NI3RNNCMAWL2V7C4LVDKPGZJUP/graph.json","fetch_events":"https://pith.science/api/pith-number/NI3RNNCMAWL2V7C4LVDKPGZJUP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NI3RNNCMAWL2V7C4LVDKPGZJUP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NI3RNNCMAWL2V7C4LVDKPGZJUP/action/storage_attestation","attest_author":"https://pith.science/pith/NI3RNNCMAWL2V7C4LVDKPGZJUP/action/author_attestation","sign_citation":"https://pith.science/pith/NI3RNNCMAWL2V7C4LVDKPGZJUP/action/citation_signature","submit_replication":"https://pith.science/pith/NI3RNNCMAWL2V7C4LVDKPGZJUP/action/replication_record"}},"created_at":"2026-07-05T06:14:33.018675+00:00","updated_at":"2026-07-05T06:14:33.018675+00:00"}