{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:XXJBPSGG25634V6POF3XTSV36M","short_pith_number":"pith:XXJBPSGG","schema_version":"1.0","canonical_sha256":"bdd217c8c6d77dbe57cf717779cabbf3350b7af09a4c2e6692ad1dd1321c8c97","source":{"kind":"arxiv","id":"2108.02040","version":2},"attestation_state":"computed","paper":{"title":"Convergence of gradient descent for learning linear neural networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.OC"],"primary_cat":"cs.LG","authors_text":"Gabin Maxime Nguegnang, Holger Rauhut, Ulrich Terstiege","submitted_at":"2021-08-04T13:10:30Z","abstract_excerpt":"We study the convergence properties of gradient descent for training deep linear neural networks, i.e., deep matrix factorizations, by extending a previous analysis for the related gradient flow. We show that under suitable conditions on the step sizes gradient descent converges to a critical point of the loss function, i.e., the square loss in this article. Furthermore, we demonstrate that for almost all initializations gradient descent converges to a global minimum in the case of two layers. In the case of three or more layers we show that gradient descent converges to a global minimum on th"},"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":"2108.02040","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-08-04T13:10:30Z","cross_cats_sorted":["math.OC"],"title_canon_sha256":"413165efe0e78318f1cb842bbad1f30314c322788b571f262ec4ef2a08d78152","abstract_canon_sha256":"2e11a8c68a9cb40a804cd0628c239c003dfcbf9ee36144a81b40bf717c0debaf"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:34:45.924080Z","signature_b64":"8e+lE3WQ+Phwo2fGzpcgRw0k2zrmcGz3Hc8FmUyM8S3FVBlDfryeleLUudnm3EAFYvSlpFLOHqOYf0k2ZyzXAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bdd217c8c6d77dbe57cf717779cabbf3350b7af09a4c2e6692ad1dd1321c8c97","last_reissued_at":"2026-07-05T03:34:45.923659Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:34:45.923659Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Convergence of gradient descent for learning linear neural networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.OC"],"primary_cat":"cs.LG","authors_text":"Gabin Maxime Nguegnang, Holger Rauhut, Ulrich Terstiege","submitted_at":"2021-08-04T13:10:30Z","abstract_excerpt":"We study the convergence properties of gradient descent for training deep linear neural networks, i.e., deep matrix factorizations, by extending a previous analysis for the related gradient flow. We show that under suitable conditions on the step sizes gradient descent converges to a critical point of the loss function, i.e., the square loss in this article. Furthermore, we demonstrate that for almost all initializations gradient descent converges to a global minimum in the case of two layers. In the case of three or more layers we show that gradient descent converges to a global minimum on th"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2108.02040","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/2108.02040/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":"2108.02040","created_at":"2026-07-05T03:34:45.923715+00:00"},{"alias_kind":"arxiv_version","alias_value":"2108.02040v2","created_at":"2026-07-05T03:34:45.923715+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2108.02040","created_at":"2026-07-05T03:34:45.923715+00:00"},{"alias_kind":"pith_short_12","alias_value":"XXJBPSGG2563","created_at":"2026-07-05T03:34:45.923715+00:00"},{"alias_kind":"pith_short_16","alias_value":"XXJBPSGG25634V6P","created_at":"2026-07-05T03:34:45.923715+00:00"},{"alias_kind":"pith_short_8","alias_value":"XXJBPSGG","created_at":"2026-07-05T03:34:45.923715+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.11664","citing_title":"A Local Polyak-Lojasiewicz and Descent Lemma of Gradient Descent For Overparametrized Linear Models","ref_index":27,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XXJBPSGG25634V6POF3XTSV36M","json":"https://pith.science/pith/XXJBPSGG25634V6POF3XTSV36M.json","graph_json":"https://pith.science/api/pith-number/XXJBPSGG25634V6POF3XTSV36M/graph.json","events_json":"https://pith.science/api/pith-number/XXJBPSGG25634V6POF3XTSV36M/events.json","paper":"https://pith.science/paper/XXJBPSGG"},"agent_actions":{"view_html":"https://pith.science/pith/XXJBPSGG25634V6POF3XTSV36M","download_json":"https://pith.science/pith/XXJBPSGG25634V6POF3XTSV36M.json","view_paper":"https://pith.science/paper/XXJBPSGG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2108.02040&json=true","fetch_graph":"https://pith.science/api/pith-number/XXJBPSGG25634V6POF3XTSV36M/graph.json","fetch_events":"https://pith.science/api/pith-number/XXJBPSGG25634V6POF3XTSV36M/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XXJBPSGG25634V6POF3XTSV36M/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XXJBPSGG25634V6POF3XTSV36M/action/storage_attestation","attest_author":"https://pith.science/pith/XXJBPSGG25634V6POF3XTSV36M/action/author_attestation","sign_citation":"https://pith.science/pith/XXJBPSGG25634V6POF3XTSV36M/action/citation_signature","submit_replication":"https://pith.science/pith/XXJBPSGG25634V6POF3XTSV36M/action/replication_record"}},"created_at":"2026-07-05T03:34:45.923715+00:00","updated_at":"2026-07-05T03:34:45.923715+00:00"}