{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2016:UWRCZMH6ZKMH4ICJTGVH23MGM6","short_pith_number":"pith:UWRCZMH6","schema_version":"1.0","canonical_sha256":"a5a22cb0feca987e204999aa7d6d8667b995f7a469cb4bb9a5793952af2b447c","source":{"kind":"arxiv","id":"1605.00405","version":2},"attestation_state":"computed","paper":{"title":"Gradient Descent Only Converges to Minimizers: Non-Isolated Critical Points and Invariant Regions","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"math.DS","authors_text":"Georgios Piliouras, Ioannis Panageas","submitted_at":"2016-05-02T09:34:19Z","abstract_excerpt":"Given a non-convex twice differentiable cost function f, we prove that the set of initial conditions so that gradient descent converges to saddle points where \\nabla^2 f has at least one strictly negative eigenvalue has (Lebesgue) measure zero, even for cost functions f with non-isolated critical points, answering an open question in [Lee, Simchowitz, Jordan, Recht, COLT2016]. Moreover, this result extends to forward-invariant convex subspaces, allowing for weak (non-globally Lipschitz) smoothness assumptions. Finally, we produce an upper bound on the allowable step-size."},"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":"1605.00405","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"math.DS","submitted_at":"2016-05-02T09:34:19Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"85f62219de5484afcabc61752c141e2729d90ff3210e998c873ae80fc12ee49f","abstract_canon_sha256":"a2beb25ac0aff437c0d7d78c6c78ecd8d0fed8e381ebe4574b81ad5d816dfae1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T01:12:48.317497Z","signature_b64":"qUhrrOPRI1kN2UaHfnKMrnshnsEMg2+hihFNdV+eX1UdMAwghiSF4t5YZqM+W/iH+9ZPJD4msrJ7X8sWPK8HCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a5a22cb0feca987e204999aa7d6d8667b995f7a469cb4bb9a5793952af2b447c","last_reissued_at":"2026-05-18T01:12:48.317147Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T01:12:48.317147Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Gradient Descent Only Converges to Minimizers: Non-Isolated Critical Points and Invariant Regions","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"math.DS","authors_text":"Georgios Piliouras, Ioannis Panageas","submitted_at":"2016-05-02T09:34:19Z","abstract_excerpt":"Given a non-convex twice differentiable cost function f, we prove that the set of initial conditions so that gradient descent converges to saddle points where \\nabla^2 f has at least one strictly negative eigenvalue has (Lebesgue) measure zero, even for cost functions f with non-isolated critical points, answering an open question in [Lee, Simchowitz, Jordan, Recht, COLT2016]. Moreover, this result extends to forward-invariant convex subspaces, allowing for weak (non-globally Lipschitz) smoothness assumptions. Finally, we produce an upper bound on the allowable step-size."},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1605.00405","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":""},"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":"1605.00405","created_at":"2026-05-18T01:12:48.317200+00:00"},{"alias_kind":"arxiv_version","alias_value":"1605.00405v2","created_at":"2026-05-18T01:12:48.317200+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1605.00405","created_at":"2026-05-18T01:12:48.317200+00:00"},{"alias_kind":"pith_short_12","alias_value":"UWRCZMH6ZKMH","created_at":"2026-05-18T12:30:46.583412+00:00"},{"alias_kind":"pith_short_16","alias_value":"UWRCZMH6ZKMH4ICJ","created_at":"2026-05-18T12:30:46.583412+00:00"},{"alias_kind":"pith_short_8","alias_value":"UWRCZMH6","created_at":"2026-05-18T12:30:46.583412+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"1906.08899","citing_title":"Limitations of Lazy Training of Two-layers Neural Networks","ref_index":29,"is_internal_anchor":true},{"citing_arxiv_id":"2307.07030","citing_title":"Accelerated Gradient Methods for Nonconvex Optimization: Escape Trajectories From Strict Saddle Points and Convergence to Local Minima","ref_index":67,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UWRCZMH6ZKMH4ICJTGVH23MGM6","json":"https://pith.science/pith/UWRCZMH6ZKMH4ICJTGVH23MGM6.json","graph_json":"https://pith.science/api/pith-number/UWRCZMH6ZKMH4ICJTGVH23MGM6/graph.json","events_json":"https://pith.science/api/pith-number/UWRCZMH6ZKMH4ICJTGVH23MGM6/events.json","paper":"https://pith.science/paper/UWRCZMH6"},"agent_actions":{"view_html":"https://pith.science/pith/UWRCZMH6ZKMH4ICJTGVH23MGM6","download_json":"https://pith.science/pith/UWRCZMH6ZKMH4ICJTGVH23MGM6.json","view_paper":"https://pith.science/paper/UWRCZMH6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1605.00405&json=true","fetch_graph":"https://pith.science/api/pith-number/UWRCZMH6ZKMH4ICJTGVH23MGM6/graph.json","fetch_events":"https://pith.science/api/pith-number/UWRCZMH6ZKMH4ICJTGVH23MGM6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UWRCZMH6ZKMH4ICJTGVH23MGM6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UWRCZMH6ZKMH4ICJTGVH23MGM6/action/storage_attestation","attest_author":"https://pith.science/pith/UWRCZMH6ZKMH4ICJTGVH23MGM6/action/author_attestation","sign_citation":"https://pith.science/pith/UWRCZMH6ZKMH4ICJTGVH23MGM6/action/citation_signature","submit_replication":"https://pith.science/pith/UWRCZMH6ZKMH4ICJTGVH23MGM6/action/replication_record"}},"created_at":"2026-05-18T01:12:48.317200+00:00","updated_at":"2026-05-18T01:12:48.317200+00:00"}