{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:FWDMYTEFNQJFUJDXJNCVP5P55T","short_pith_number":"pith:FWDMYTEF","schema_version":"1.0","canonical_sha256":"2d86cc4c856c125a24774b4557f5fdecf5ddcfe3a70bc73a6498cbf20a8861c1","source":{"kind":"arxiv","id":"2103.10427","version":4},"attestation_state":"computed","paper":{"title":"The Low-Rank Simplicity Bias in Deep Networks","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Brian Cheung, Hossein Mobahi, Minyoung Huh, Phillip Isola, Pulkit Agrawal, Richard Zhang","submitted_at":"2021-03-18T17:58:02Z","abstract_excerpt":"Modern deep neural networks are highly over-parameterized compared to the data on which they are trained, yet they often generalize remarkably well. A flurry of recent work has asked: why do deep networks not overfit to their training data? In this work, we make a series of empirical observations that investigate and extend the hypothesis that deeper networks are inductively biased to find solutions with lower effective rank embeddings. We conjecture that this bias exists because the volume of functions that maps to low effective rank embedding increases with depth. We show empirically that ou"},"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":"2103.10427","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2021-03-18T17:58:02Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"1b891897a4a624cb34cd64b3573d2ccf230367f8311fceb0520d5ac4eb7c7f0b","abstract_canon_sha256":"7c3085d6ca3f482418ef2fff93229942656cc63f559c3890f88a1dace3a26113"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:53:46.189498Z","signature_b64":"A5eXEBMkimi1aR02Py87ER2DcTA54lCexO2GkxB+r2t72IWo4Kr8noRUs5rwQV3oNUGhFk48ShoQVomqXI/gCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2d86cc4c856c125a24774b4557f5fdecf5ddcfe3a70bc73a6498cbf20a8861c1","last_reissued_at":"2026-07-05T05:53:46.189031Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:53:46.189031Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Low-Rank Simplicity Bias in Deep Networks","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Brian Cheung, Hossein Mobahi, Minyoung Huh, Phillip Isola, Pulkit Agrawal, Richard Zhang","submitted_at":"2021-03-18T17:58:02Z","abstract_excerpt":"Modern deep neural networks are highly over-parameterized compared to the data on which they are trained, yet they often generalize remarkably well. A flurry of recent work has asked: why do deep networks not overfit to their training data? In this work, we make a series of empirical observations that investigate and extend the hypothesis that deeper networks are inductively biased to find solutions with lower effective rank embeddings. We conjecture that this bias exists because the volume of functions that maps to low effective rank embedding increases with depth. We show empirically that ou"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2103.10427","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/2103.10427/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":"2103.10427","created_at":"2026-07-05T05:53:46.189088+00:00"},{"alias_kind":"arxiv_version","alias_value":"2103.10427v4","created_at":"2026-07-05T05:53:46.189088+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2103.10427","created_at":"2026-07-05T05:53:46.189088+00:00"},{"alias_kind":"pith_short_12","alias_value":"FWDMYTEFNQJF","created_at":"2026-07-05T05:53:46.189088+00:00"},{"alias_kind":"pith_short_16","alias_value":"FWDMYTEFNQJFUJDX","created_at":"2026-07-05T05:53:46.189088+00:00"},{"alias_kind":"pith_short_8","alias_value":"FWDMYTEF","created_at":"2026-07-05T05:53:46.189088+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11123","citing_title":"Overcoming Rank Collapse in Feedback Alignment","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01412","citing_title":"GPTQ-intrinsic LoRA: A Near-optimal Algorithm for Low-precision Quantization with Low-rank Adaptation","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29885","citing_title":"Open Problem: Separating Geometric and Algorithmic Compression via Cayley-Table Completion","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2503.03206","citing_title":"An Analytical Theory of Spectral Bias in the Learning Dynamics of Diffusion Models","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2508.04853","citing_title":"Provable Post-Training Quantization: Theoretical Analysis of OPTQ and Qronos","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2509.18993","citing_title":"CR-Net: Scaling Parameter-Efficient Training with Cross-Layer Low-Rank Structure","ref_index":29,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FWDMYTEFNQJFUJDXJNCVP5P55T","json":"https://pith.science/pith/FWDMYTEFNQJFUJDXJNCVP5P55T.json","graph_json":"https://pith.science/api/pith-number/FWDMYTEFNQJFUJDXJNCVP5P55T/graph.json","events_json":"https://pith.science/api/pith-number/FWDMYTEFNQJFUJDXJNCVP5P55T/events.json","paper":"https://pith.science/paper/FWDMYTEF"},"agent_actions":{"view_html":"https://pith.science/pith/FWDMYTEFNQJFUJDXJNCVP5P55T","download_json":"https://pith.science/pith/FWDMYTEFNQJFUJDXJNCVP5P55T.json","view_paper":"https://pith.science/paper/FWDMYTEF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2103.10427&json=true","fetch_graph":"https://pith.science/api/pith-number/FWDMYTEFNQJFUJDXJNCVP5P55T/graph.json","fetch_events":"https://pith.science/api/pith-number/FWDMYTEFNQJFUJDXJNCVP5P55T/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FWDMYTEFNQJFUJDXJNCVP5P55T/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FWDMYTEFNQJFUJDXJNCVP5P55T/action/storage_attestation","attest_author":"https://pith.science/pith/FWDMYTEFNQJFUJDXJNCVP5P55T/action/author_attestation","sign_citation":"https://pith.science/pith/FWDMYTEFNQJFUJDXJNCVP5P55T/action/citation_signature","submit_replication":"https://pith.science/pith/FWDMYTEFNQJFUJDXJNCVP5P55T/action/replication_record"}},"created_at":"2026-07-05T05:53:46.189088+00:00","updated_at":"2026-07-05T05:53:46.189088+00:00"}