{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:N7UCSJ6OMZ63OPSZQDO3Q6KR33","short_pith_number":"pith:N7UCSJ6O","schema_version":"1.0","canonical_sha256":"6fe82927ce667db73e5980ddb87951deeebecd9393a5d372aefc44c63c3302af","source":{"kind":"arxiv","id":"2011.09468","version":4},"attestation_state":"computed","paper":{"title":"Gradient Starvation: A Learning Proclivity in Neural Networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.DS","stat.ML"],"primary_cat":"cs.LG","authors_text":"Aaron Courville, Doina Precup, Guillaume Lajoie, Mohammad Pezeshki, S\\'ekou-Oumar Kaba, Yoshua Bengio","submitted_at":"2020-11-18T18:52:08Z","abstract_excerpt":"We identify and formalize a fundamental gradient descent phenomenon resulting in a learning proclivity in over-parameterized neural networks. Gradient Starvation arises when cross-entropy loss is minimized by capturing only a subset of features relevant for the task, despite the presence of other predictive features that fail to be discovered. This work provides a theoretical explanation for the emergence of such feature imbalance in neural networks. Using tools from Dynamical Systems theory, we identify simple properties of learning dynamics during gradient descent that lead to this imbalance"},"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":"2011.09468","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2020-11-18T18:52:08Z","cross_cats_sorted":["math.DS","stat.ML"],"title_canon_sha256":"61d9aaeffe0e59c41d88f6e1f440143b14c827fe7c4a32488414ec720b5c4bb7","abstract_canon_sha256":"81e43499cde38766d8fea1d83bf002caef77252ee2bc2544a432a98e7d7b5ee3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:34:42.427828Z","signature_b64":"I8zK5MteUmXWvAQnT838TdhupfV0bqTMx+CUDf6T8XF/WhQjYOTAVegNrw1J1YRiAzJRB3LmvnURYtoSAIUEDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6fe82927ce667db73e5980ddb87951deeebecd9393a5d372aefc44c63c3302af","last_reissued_at":"2026-07-05T03:34:42.427393Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:34:42.427393Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Gradient Starvation: A Learning Proclivity in Neural Networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.DS","stat.ML"],"primary_cat":"cs.LG","authors_text":"Aaron Courville, Doina Precup, Guillaume Lajoie, Mohammad Pezeshki, S\\'ekou-Oumar Kaba, Yoshua Bengio","submitted_at":"2020-11-18T18:52:08Z","abstract_excerpt":"We identify and formalize a fundamental gradient descent phenomenon resulting in a learning proclivity in over-parameterized neural networks. Gradient Starvation arises when cross-entropy loss is minimized by capturing only a subset of features relevant for the task, despite the presence of other predictive features that fail to be discovered. This work provides a theoretical explanation for the emergence of such feature imbalance in neural networks. Using tools from Dynamical Systems theory, we identify simple properties of learning dynamics during gradient descent that lead to this imbalance"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2011.09468","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/2011.09468/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":"2011.09468","created_at":"2026-07-05T03:34:42.427453+00:00"},{"alias_kind":"arxiv_version","alias_value":"2011.09468v4","created_at":"2026-07-05T03:34:42.427453+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2011.09468","created_at":"2026-07-05T03:34:42.427453+00:00"},{"alias_kind":"pith_short_12","alias_value":"N7UCSJ6OMZ63","created_at":"2026-07-05T03:34:42.427453+00:00"},{"alias_kind":"pith_short_16","alias_value":"N7UCSJ6OMZ63OPSZ","created_at":"2026-07-05T03:34:42.427453+00:00"},{"alias_kind":"pith_short_8","alias_value":"N7UCSJ6O","created_at":"2026-07-05T03:34:42.427453+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.20771","citing_title":"Cumulative Meta-Learning from Active Learning Queries for Robustness to Spurious Correlations","ref_index":71,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20732","citing_title":"Deep Attention Reweighting: Post-Hoc Attention-Based Feature Aggregation in CNNs for Disentangling Core and Spurious Features under Spurious Correlations","ref_index":56,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/N7UCSJ6OMZ63OPSZQDO3Q6KR33","json":"https://pith.science/pith/N7UCSJ6OMZ63OPSZQDO3Q6KR33.json","graph_json":"https://pith.science/api/pith-number/N7UCSJ6OMZ63OPSZQDO3Q6KR33/graph.json","events_json":"https://pith.science/api/pith-number/N7UCSJ6OMZ63OPSZQDO3Q6KR33/events.json","paper":"https://pith.science/paper/N7UCSJ6O"},"agent_actions":{"view_html":"https://pith.science/pith/N7UCSJ6OMZ63OPSZQDO3Q6KR33","download_json":"https://pith.science/pith/N7UCSJ6OMZ63OPSZQDO3Q6KR33.json","view_paper":"https://pith.science/paper/N7UCSJ6O","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2011.09468&json=true","fetch_graph":"https://pith.science/api/pith-number/N7UCSJ6OMZ63OPSZQDO3Q6KR33/graph.json","fetch_events":"https://pith.science/api/pith-number/N7UCSJ6OMZ63OPSZQDO3Q6KR33/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/N7UCSJ6OMZ63OPSZQDO3Q6KR33/action/timestamp_anchor","attest_storage":"https://pith.science/pith/N7UCSJ6OMZ63OPSZQDO3Q6KR33/action/storage_attestation","attest_author":"https://pith.science/pith/N7UCSJ6OMZ63OPSZQDO3Q6KR33/action/author_attestation","sign_citation":"https://pith.science/pith/N7UCSJ6OMZ63OPSZQDO3Q6KR33/action/citation_signature","submit_replication":"https://pith.science/pith/N7UCSJ6OMZ63OPSZQDO3Q6KR33/action/replication_record"}},"created_at":"2026-07-05T03:34:42.427453+00:00","updated_at":"2026-07-05T03:34:42.427453+00:00"}