{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:52QMVF6BIVA3GFSBUQSQZX37TT","short_pith_number":"pith:52QMVF6B","schema_version":"1.0","canonical_sha256":"eea0ca97c14541b31641a4250cdf7f9cd1f07c8e59897916ff291c9576e5c360","source":{"kind":"arxiv","id":"2001.04413","version":6},"attestation_state":"computed","paper":{"title":"Backward Feature Correction: How Deep Learning Performs Deep (Hierarchical) Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.DS","cs.NE","math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Yuanzhi Li, Zeyuan Allen-Zhu","submitted_at":"2020-01-13T17:28:29Z","abstract_excerpt":"Deep learning is also known as hierarchical learning, where the learner _learns_ to represent a complicated target function by decomposing it into a sequence of simpler functions to reduce sample and time complexity. This paper formally analyzes how multi-layer neural networks can perform such hierarchical learning _efficiently_ and _automatically_ by SGD on the training objective.\n  On the conceptual side, we present a theoretical characterizations of how certain types of deep (i.e. super-constant layer) neural networks can still be sample and time efficiently trained on some hierarchical tas"},"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":"2001.04413","kind":"arxiv","version":6},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2020-01-13T17:28:29Z","cross_cats_sorted":["cs.DS","cs.NE","math.OC","stat.ML"],"title_canon_sha256":"88ec60b3cb02e31cd93dad67794af9e3cbe80a94e14b02ab58be16a65183158e","abstract_canon_sha256":"ebca7244159be0fc25d41611e366d1cf976867491eda13351a8a96c42ad77dc9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:28:29.249523Z","signature_b64":"aV+NjCcc795/5+en3MPScbqqx9atZ+Wi2aGrEeFoZxprP3N85WS7ODmOsa8oQDwVmJiq+DgVbKFwyz4JTIDKDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"eea0ca97c14541b31641a4250cdf7f9cd1f07c8e59897916ff291c9576e5c360","last_reissued_at":"2026-07-05T06:28:29.249053Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:28:29.249053Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Backward Feature Correction: How Deep Learning Performs Deep (Hierarchical) Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.DS","cs.NE","math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Yuanzhi Li, Zeyuan Allen-Zhu","submitted_at":"2020-01-13T17:28:29Z","abstract_excerpt":"Deep learning is also known as hierarchical learning, where the learner _learns_ to represent a complicated target function by decomposing it into a sequence of simpler functions to reduce sample and time complexity. This paper formally analyzes how multi-layer neural networks can perform such hierarchical learning _efficiently_ and _automatically_ by SGD on the training objective.\n  On the conceptual side, we present a theoretical characterizations of how certain types of deep (i.e. super-constant layer) neural networks can still be sample and time efficiently trained on some hierarchical tas"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2001.04413","kind":"arxiv","version":6},"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/2001.04413/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":"2001.04413","created_at":"2026-07-05T06:28:29.249106+00:00"},{"alias_kind":"arxiv_version","alias_value":"2001.04413v6","created_at":"2026-07-05T06:28:29.249106+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2001.04413","created_at":"2026-07-05T06:28:29.249106+00:00"},{"alias_kind":"pith_short_12","alias_value":"52QMVF6BIVA3","created_at":"2026-07-05T06:28:29.249106+00:00"},{"alias_kind":"pith_short_16","alias_value":"52QMVF6BIVA3GFSB","created_at":"2026-07-05T06:28:29.249106+00:00"},{"alias_kind":"pith_short_8","alias_value":"52QMVF6B","created_at":"2026-07-05T06:28:29.249106+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.10089","citing_title":"A Theory on Flow Matching with Neural Networks","ref_index":261,"is_internal_anchor":false},{"citing_arxiv_id":"2401.01335","citing_title":"Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models","ref_index":242,"is_internal_anchor":false},{"citing_arxiv_id":"2106.09685","citing_title":"LoRA: Low-Rank Adaptation of Large Language Models","ref_index":3,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/52QMVF6BIVA3GFSBUQSQZX37TT","json":"https://pith.science/pith/52QMVF6BIVA3GFSBUQSQZX37TT.json","graph_json":"https://pith.science/api/pith-number/52QMVF6BIVA3GFSBUQSQZX37TT/graph.json","events_json":"https://pith.science/api/pith-number/52QMVF6BIVA3GFSBUQSQZX37TT/events.json","paper":"https://pith.science/paper/52QMVF6B"},"agent_actions":{"view_html":"https://pith.science/pith/52QMVF6BIVA3GFSBUQSQZX37TT","download_json":"https://pith.science/pith/52QMVF6BIVA3GFSBUQSQZX37TT.json","view_paper":"https://pith.science/paper/52QMVF6B","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2001.04413&json=true","fetch_graph":"https://pith.science/api/pith-number/52QMVF6BIVA3GFSBUQSQZX37TT/graph.json","fetch_events":"https://pith.science/api/pith-number/52QMVF6BIVA3GFSBUQSQZX37TT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/52QMVF6BIVA3GFSBUQSQZX37TT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/52QMVF6BIVA3GFSBUQSQZX37TT/action/storage_attestation","attest_author":"https://pith.science/pith/52QMVF6BIVA3GFSBUQSQZX37TT/action/author_attestation","sign_citation":"https://pith.science/pith/52QMVF6BIVA3GFSBUQSQZX37TT/action/citation_signature","submit_replication":"https://pith.science/pith/52QMVF6BIVA3GFSBUQSQZX37TT/action/replication_record"}},"created_at":"2026-07-05T06:28:29.249106+00:00","updated_at":"2026-07-05T06:28:29.249106+00:00"}