{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:K4ERKXRH3IJUR4W5WFKERKKG4A","short_pith_number":"pith:K4ERKXRH","schema_version":"1.0","canonical_sha256":"5709155e27da1348f2ddb15448a946e02aaf79d9e4d36d1d0c07b37f92d8d65a","source":{"kind":"arxiv","id":"2202.09028","version":9},"attestation_state":"computed","paper":{"title":"On the Implicit Bias Towards Minimal Depth of Deep Neural Networks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Ido Ben-Shaul, Liane Galanti, Tomer Galanti","submitted_at":"2022-02-18T05:21:28Z","abstract_excerpt":"Recent results in the literature suggest that the penultimate (second-to-last) layer representations of neural networks that are trained for classification exhibit a clustering property called neural collapse (NC). We study the implicit bias of stochastic gradient descent (SGD) in favor of low-depth solutions when training deep neural networks. We characterize a notion of effective depth that measures the first layer for which sample embeddings are separable using the nearest-class center classifier. Furthermore, we hypothesize and empirically show that SGD implicitly selects neural networks o"},"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":"2202.09028","kind":"arxiv","version":9},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2022-02-18T05:21:28Z","cross_cats_sorted":[],"title_canon_sha256":"9ae2c7e48f67a90be296199ecb352af15f05ca262125569dacd0df16d28aaf3f","abstract_canon_sha256":"fbd6f9032173744ee4a2b552a24b7c3251a11d5b26918aedc37456962f603bbe"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:01:37.478543Z","signature_b64":"d9sAQlJEVsTwk+EuqrKRhCEEHNnf6zK462TlBcLZibEgRiJ7E/4MZqWef9QGTG9f2ygVb7saf+QEb041ua9TBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5709155e27da1348f2ddb15448a946e02aaf79d9e4d36d1d0c07b37f92d8d65a","last_reissued_at":"2026-07-05T05:01:37.478119Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:01:37.478119Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On the Implicit Bias Towards Minimal Depth of Deep Neural Networks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Ido Ben-Shaul, Liane Galanti, Tomer Galanti","submitted_at":"2022-02-18T05:21:28Z","abstract_excerpt":"Recent results in the literature suggest that the penultimate (second-to-last) layer representations of neural networks that are trained for classification exhibit a clustering property called neural collapse (NC). We study the implicit bias of stochastic gradient descent (SGD) in favor of low-depth solutions when training deep neural networks. We characterize a notion of effective depth that measures the first layer for which sample embeddings are separable using the nearest-class center classifier. Furthermore, we hypothesize and empirically show that SGD implicitly selects neural networks o"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2202.09028","kind":"arxiv","version":9},"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/2202.09028/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":"2202.09028","created_at":"2026-07-05T05:01:37.478175+00:00"},{"alias_kind":"arxiv_version","alias_value":"2202.09028v9","created_at":"2026-07-05T05:01:37.478175+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2202.09028","created_at":"2026-07-05T05:01:37.478175+00:00"},{"alias_kind":"pith_short_12","alias_value":"K4ERKXRH3IJU","created_at":"2026-07-05T05:01:37.478175+00:00"},{"alias_kind":"pith_short_16","alias_value":"K4ERKXRH3IJUR4W5","created_at":"2026-07-05T05:01:37.478175+00:00"},{"alias_kind":"pith_short_8","alias_value":"K4ERKXRH","created_at":"2026-07-05T05:01:37.478175+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.19680","citing_title":"Feature learning is decoupled from generalization in high capacity neural networks","ref_index":25,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/K4ERKXRH3IJUR4W5WFKERKKG4A","json":"https://pith.science/pith/K4ERKXRH3IJUR4W5WFKERKKG4A.json","graph_json":"https://pith.science/api/pith-number/K4ERKXRH3IJUR4W5WFKERKKG4A/graph.json","events_json":"https://pith.science/api/pith-number/K4ERKXRH3IJUR4W5WFKERKKG4A/events.json","paper":"https://pith.science/paper/K4ERKXRH"},"agent_actions":{"view_html":"https://pith.science/pith/K4ERKXRH3IJUR4W5WFKERKKG4A","download_json":"https://pith.science/pith/K4ERKXRH3IJUR4W5WFKERKKG4A.json","view_paper":"https://pith.science/paper/K4ERKXRH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2202.09028&json=true","fetch_graph":"https://pith.science/api/pith-number/K4ERKXRH3IJUR4W5WFKERKKG4A/graph.json","fetch_events":"https://pith.science/api/pith-number/K4ERKXRH3IJUR4W5WFKERKKG4A/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/K4ERKXRH3IJUR4W5WFKERKKG4A/action/timestamp_anchor","attest_storage":"https://pith.science/pith/K4ERKXRH3IJUR4W5WFKERKKG4A/action/storage_attestation","attest_author":"https://pith.science/pith/K4ERKXRH3IJUR4W5WFKERKKG4A/action/author_attestation","sign_citation":"https://pith.science/pith/K4ERKXRH3IJUR4W5WFKERKKG4A/action/citation_signature","submit_replication":"https://pith.science/pith/K4ERKXRH3IJUR4W5WFKERKKG4A/action/replication_record"}},"created_at":"2026-07-05T05:01:37.478175+00:00","updated_at":"2026-07-05T05:01:37.478175+00:00"}