{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:FFXQPV5FJBU6EJJP4255LWIYWY","short_pith_number":"pith:FFXQPV5F","schema_version":"1.0","canonical_sha256":"296f07d7a54869e2252fe6bbd5d918b63c2c9df067093a309f6eb567f210deae","source":{"kind":"arxiv","id":"2209.15055","version":4},"attestation_state":"computed","paper":{"title":"Implicit Bias of Large Depth Networks: a Notion of Rank for Nonlinear Functions","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"stat.ML","authors_text":"Arthur Jacot","submitted_at":"2022-09-29T18:57:51Z","abstract_excerpt":"We show that the representation cost of fully connected neural networks with homogeneous nonlinearities - which describes the implicit bias in function space of networks with $L_2$-regularization or with losses such as the cross-entropy - converges as the depth of the network goes to infinity to a notion of rank over nonlinear functions. We then inquire under which conditions the global minima of the loss recover the `true' rank of the data: we show that for too large depths the global minimum will be approximately rank 1 (underestimating the rank); we then argue that there is a range of depth"},"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":"2209.15055","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2022-09-29T18:57:51Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"160a23d090b251ed33bf5bf6859cfa186e5534fc855611eace4f19b9b8c5878b","abstract_canon_sha256":"6183cdaf8593ba04dbb2d468fdb6da4922014ba6a3315196a54c294885c917fb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:54:03.604409Z","signature_b64":"1aOWDlGsViPGGf0A/yAowXPpt437dvBz97/u5L3y7HXKkh/xOExNzgn1O93+Eqo3F8qfYwJsHqB/HvuTCAGZBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"296f07d7a54869e2252fe6bbd5d918b63c2c9df067093a309f6eb567f210deae","last_reissued_at":"2026-07-05T05:54:03.604014Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:54:03.604014Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Implicit Bias of Large Depth Networks: a Notion of Rank for Nonlinear Functions","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"stat.ML","authors_text":"Arthur Jacot","submitted_at":"2022-09-29T18:57:51Z","abstract_excerpt":"We show that the representation cost of fully connected neural networks with homogeneous nonlinearities - which describes the implicit bias in function space of networks with $L_2$-regularization or with losses such as the cross-entropy - converges as the depth of the network goes to infinity to a notion of rank over nonlinear functions. We then inquire under which conditions the global minima of the loss recover the `true' rank of the data: we show that for too large depths the global minimum will be approximately rank 1 (underestimating the rank); we then argue that there is a range of depth"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2209.15055","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/2209.15055/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":"2209.15055","created_at":"2026-07-05T05:54:03.604070+00:00"},{"alias_kind":"arxiv_version","alias_value":"2209.15055v4","created_at":"2026-07-05T05:54:03.604070+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2209.15055","created_at":"2026-07-05T05:54:03.604070+00:00"},{"alias_kind":"pith_short_12","alias_value":"FFXQPV5FJBU6","created_at":"2026-07-05T05:54:03.604070+00:00"},{"alias_kind":"pith_short_16","alias_value":"FFXQPV5FJBU6EJJP","created_at":"2026-07-05T05:54:03.604070+00:00"},{"alias_kind":"pith_short_8","alias_value":"FFXQPV5F","created_at":"2026-07-05T05:54:03.604070+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.11370","citing_title":"Understanding Nonlinear Implicit Bias via Region Counts in Input Space","ref_index":2018,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FFXQPV5FJBU6EJJP4255LWIYWY","json":"https://pith.science/pith/FFXQPV5FJBU6EJJP4255LWIYWY.json","graph_json":"https://pith.science/api/pith-number/FFXQPV5FJBU6EJJP4255LWIYWY/graph.json","events_json":"https://pith.science/api/pith-number/FFXQPV5FJBU6EJJP4255LWIYWY/events.json","paper":"https://pith.science/paper/FFXQPV5F"},"agent_actions":{"view_html":"https://pith.science/pith/FFXQPV5FJBU6EJJP4255LWIYWY","download_json":"https://pith.science/pith/FFXQPV5FJBU6EJJP4255LWIYWY.json","view_paper":"https://pith.science/paper/FFXQPV5F","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2209.15055&json=true","fetch_graph":"https://pith.science/api/pith-number/FFXQPV5FJBU6EJJP4255LWIYWY/graph.json","fetch_events":"https://pith.science/api/pith-number/FFXQPV5FJBU6EJJP4255LWIYWY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FFXQPV5FJBU6EJJP4255LWIYWY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FFXQPV5FJBU6EJJP4255LWIYWY/action/storage_attestation","attest_author":"https://pith.science/pith/FFXQPV5FJBU6EJJP4255LWIYWY/action/author_attestation","sign_citation":"https://pith.science/pith/FFXQPV5FJBU6EJJP4255LWIYWY/action/citation_signature","submit_replication":"https://pith.science/pith/FFXQPV5FJBU6EJJP4255LWIYWY/action/replication_record"}},"created_at":"2026-07-05T05:54:03.604070+00:00","updated_at":"2026-07-05T05:54:03.604070+00:00"}