{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:4ZALQAHJ6MS3W7GNOGP5ZYCRH7","short_pith_number":"pith:4ZALQAHJ","schema_version":"1.0","canonical_sha256":"e640b800e9f325bb7ccd719fdce0513ff81a50cb94e82c791e5b48ece137f564","source":{"kind":"arxiv","id":"1901.09149","version":2},"attestation_state":"computed","paper":{"title":"Escaping Saddle Points with Adaptive Gradient Methods","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Matthew Staib, Sanjiv Kumar, Sashank J. Reddi, Satyen Kale, Suvrit Sra","submitted_at":"2019-01-26T03:12:14Z","abstract_excerpt":"Adaptive methods such as Adam and RMSProp are widely used in deep learning but are not well understood. In this paper, we seek a crisp, clean and precise characterization of their behavior in nonconvex settings. To this end, we first provide a novel view of adaptive methods as preconditioned SGD, where the preconditioner is estimated in an online manner. By studying the preconditioner on its own, we elucidate its purpose: it rescales the stochastic gradient noise to be isotropic near stationary points, which helps escape saddle points. Furthermore, we show that adaptive methods can efficiently"},"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":"1901.09149","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2019-01-26T03:12:14Z","cross_cats_sorted":["math.OC","stat.ML"],"title_canon_sha256":"10a6c0db9bcbb7fa04b899f99b117f97c55a4c27a1e3a09bb15e8a9d236f7a5e","abstract_canon_sha256":"13d3ea94ed91ca6a6fa07f9ac831e3699a27bb2b78f18db5da4c1a903572844a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T00:37:43.999098Z","signature_b64":"o5sWK2n0drzMz8cN8bZ23RY0cmo7vuhQLNIeWX5A/C9mem1H0r1nHNKUa0+YeC49AL8XDkls1DvQfjOQISPVBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e640b800e9f325bb7ccd719fdce0513ff81a50cb94e82c791e5b48ece137f564","last_reissued_at":"2026-07-05T00:37:43.998704Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T00:37:43.998704Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Escaping Saddle Points with Adaptive Gradient Methods","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Matthew Staib, Sanjiv Kumar, Sashank J. Reddi, Satyen Kale, Suvrit Sra","submitted_at":"2019-01-26T03:12:14Z","abstract_excerpt":"Adaptive methods such as Adam and RMSProp are widely used in deep learning but are not well understood. In this paper, we seek a crisp, clean and precise characterization of their behavior in nonconvex settings. To this end, we first provide a novel view of adaptive methods as preconditioned SGD, where the preconditioner is estimated in an online manner. By studying the preconditioner on its own, we elucidate its purpose: it rescales the stochastic gradient noise to be isotropic near stationary points, which helps escape saddle points. Furthermore, we show that adaptive methods can efficiently"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1901.09149","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/1901.09149/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":"1901.09149","created_at":"2026-07-05T00:37:43.998759+00:00"},{"alias_kind":"arxiv_version","alias_value":"1901.09149v2","created_at":"2026-07-05T00:37:43.998759+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1901.09149","created_at":"2026-07-05T00:37:43.998759+00:00"},{"alias_kind":"pith_short_12","alias_value":"4ZALQAHJ6MS3","created_at":"2026-07-05T00:37:43.998759+00:00"},{"alias_kind":"pith_short_16","alias_value":"4ZALQAHJ6MS3W7GN","created_at":"2026-07-05T00:37:43.998759+00:00"},{"alias_kind":"pith_short_8","alias_value":"4ZALQAHJ","created_at":"2026-07-05T00:37:43.998759+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2003.00295","citing_title":"Adaptive Federated Optimization","ref_index":200,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09331","citing_title":"Dimension-Free Saddle-Point Escape in Muon","ref_index":30,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4ZALQAHJ6MS3W7GNOGP5ZYCRH7","json":"https://pith.science/pith/4ZALQAHJ6MS3W7GNOGP5ZYCRH7.json","graph_json":"https://pith.science/api/pith-number/4ZALQAHJ6MS3W7GNOGP5ZYCRH7/graph.json","events_json":"https://pith.science/api/pith-number/4ZALQAHJ6MS3W7GNOGP5ZYCRH7/events.json","paper":"https://pith.science/paper/4ZALQAHJ"},"agent_actions":{"view_html":"https://pith.science/pith/4ZALQAHJ6MS3W7GNOGP5ZYCRH7","download_json":"https://pith.science/pith/4ZALQAHJ6MS3W7GNOGP5ZYCRH7.json","view_paper":"https://pith.science/paper/4ZALQAHJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1901.09149&json=true","fetch_graph":"https://pith.science/api/pith-number/4ZALQAHJ6MS3W7GNOGP5ZYCRH7/graph.json","fetch_events":"https://pith.science/api/pith-number/4ZALQAHJ6MS3W7GNOGP5ZYCRH7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4ZALQAHJ6MS3W7GNOGP5ZYCRH7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4ZALQAHJ6MS3W7GNOGP5ZYCRH7/action/storage_attestation","attest_author":"https://pith.science/pith/4ZALQAHJ6MS3W7GNOGP5ZYCRH7/action/author_attestation","sign_citation":"https://pith.science/pith/4ZALQAHJ6MS3W7GNOGP5ZYCRH7/action/citation_signature","submit_replication":"https://pith.science/pith/4ZALQAHJ6MS3W7GNOGP5ZYCRH7/action/replication_record"}},"created_at":"2026-07-05T00:37:43.998759+00:00","updated_at":"2026-07-05T00:37:43.998759+00:00"}