{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:SAGL45HCVOC5WQHAO4TKU4HRL3","short_pith_number":"pith:SAGL45HC","schema_version":"1.0","canonical_sha256":"900cbe74e2ab85db40e07726aa70f15edd0404f44a0babc4bcd60aa729aa34be","source":{"kind":"arxiv","id":"2210.03294","version":2},"attestation_state":"computed","paper":{"title":"Understanding Edge-of-Stability Training Dynamics with a Minimalist Example","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Mo Zhou, Rong Ge, Xiang Wang, Xingyu Zhu, Zixuan Wang","submitted_at":"2022-10-07T02:57:05Z","abstract_excerpt":"Recently, researchers observed that gradient descent for deep neural networks operates in an ``edge-of-stability'' (EoS) regime: the sharpness (maximum eigenvalue of the Hessian) is often larger than stability threshold $2/\\eta$ (where $\\eta$ is the step size). Despite this, the loss oscillates and converges in the long run, and the sharpness at the end is just slightly below $2/\\eta$. While many other well-understood nonconvex objectives such as matrix factorization or two-layer networks can also converge despite large sharpness, there is often a larger gap between sharpness of the endpoint a"},"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":"2210.03294","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2022-10-07T02:57:05Z","cross_cats_sorted":["math.OC","stat.ML"],"title_canon_sha256":"1fb93e5ce38c5074af2f99c3362460ae706d69138fb1e234a80fa99b181ed92f","abstract_canon_sha256":"c87eb24eac2e2b094854c029e075aeceb8ec7c57b5714cc9b91f47f638687481"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:43:58.429210Z","signature_b64":"5pzet+ooRUtaRh+JMCu+d+Qkt+DoZ1WKdKPoZJedLX1TWfsYprUjHc5JKiBNbt79twnmgU0smKa7bq4MfaybDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"900cbe74e2ab85db40e07726aa70f15edd0404f44a0babc4bcd60aa729aa34be","last_reissued_at":"2026-07-05T05:43:58.428703Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:43:58.428703Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding Edge-of-Stability Training Dynamics with a Minimalist Example","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Mo Zhou, Rong Ge, Xiang Wang, Xingyu Zhu, Zixuan Wang","submitted_at":"2022-10-07T02:57:05Z","abstract_excerpt":"Recently, researchers observed that gradient descent for deep neural networks operates in an ``edge-of-stability'' (EoS) regime: the sharpness (maximum eigenvalue of the Hessian) is often larger than stability threshold $2/\\eta$ (where $\\eta$ is the step size). Despite this, the loss oscillates and converges in the long run, and the sharpness at the end is just slightly below $2/\\eta$. While many other well-understood nonconvex objectives such as matrix factorization or two-layer networks can also converge despite large sharpness, there is often a larger gap between sharpness of the endpoint a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2210.03294","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/2210.03294/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":"2210.03294","created_at":"2026-07-05T05:43:58.428762+00:00"},{"alias_kind":"arxiv_version","alias_value":"2210.03294v2","created_at":"2026-07-05T05:43:58.428762+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2210.03294","created_at":"2026-07-05T05:43:58.428762+00:00"},{"alias_kind":"pith_short_12","alias_value":"SAGL45HCVOC5","created_at":"2026-07-05T05:43:58.428762+00:00"},{"alias_kind":"pith_short_16","alias_value":"SAGL45HCVOC5WQHA","created_at":"2026-07-05T05:43:58.428762+00:00"},{"alias_kind":"pith_short_8","alias_value":"SAGL45HC","created_at":"2026-07-05T05:43:58.428762+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.05326","citing_title":"Gradient descent at the Edge of Stability: free energy model and kinetic description of the two-layer network","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19740","citing_title":"Generalization at the Edge of Stability","ref_index":79,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SAGL45HCVOC5WQHAO4TKU4HRL3","json":"https://pith.science/pith/SAGL45HCVOC5WQHAO4TKU4HRL3.json","graph_json":"https://pith.science/api/pith-number/SAGL45HCVOC5WQHAO4TKU4HRL3/graph.json","events_json":"https://pith.science/api/pith-number/SAGL45HCVOC5WQHAO4TKU4HRL3/events.json","paper":"https://pith.science/paper/SAGL45HC"},"agent_actions":{"view_html":"https://pith.science/pith/SAGL45HCVOC5WQHAO4TKU4HRL3","download_json":"https://pith.science/pith/SAGL45HCVOC5WQHAO4TKU4HRL3.json","view_paper":"https://pith.science/paper/SAGL45HC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2210.03294&json=true","fetch_graph":"https://pith.science/api/pith-number/SAGL45HCVOC5WQHAO4TKU4HRL3/graph.json","fetch_events":"https://pith.science/api/pith-number/SAGL45HCVOC5WQHAO4TKU4HRL3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SAGL45HCVOC5WQHAO4TKU4HRL3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SAGL45HCVOC5WQHAO4TKU4HRL3/action/storage_attestation","attest_author":"https://pith.science/pith/SAGL45HCVOC5WQHAO4TKU4HRL3/action/author_attestation","sign_citation":"https://pith.science/pith/SAGL45HCVOC5WQHAO4TKU4HRL3/action/citation_signature","submit_replication":"https://pith.science/pith/SAGL45HCVOC5WQHAO4TKU4HRL3/action/replication_record"}},"created_at":"2026-07-05T05:43:58.428762+00:00","updated_at":"2026-07-05T05:43:58.428762+00:00"}