{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:XPM3SDEEKCQWY44JLNQ7JW26U5","short_pith_number":"pith:XPM3SDEE","schema_version":"1.0","canonical_sha256":"bbd9b90c8450a16c73895b61f4db5ea74eb659870166f9b4124f36bfd778de4b","source":{"kind":"arxiv","id":"2403.03950","version":1},"attestation_state":"computed","paper":{"title":"Stop Regressing: Training Value Functions via Classification for Scalable Deep RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Adrien Ali Ta\\\"iga, Aleksandra Faust, Alex Irpan, Aviral Kumar, Jesse Farebrother, Jordi Orbay, Pablo Samuel Castro, Quan Vuong, Rishabh Agarwal, Sergey Levine, Ted Xiao, Yevgen Chebotar","submitted_at":"2024-03-06T18:55:47Z","abstract_excerpt":"Value functions are a central component of deep reinforcement learning (RL). These functions, parameterized by neural networks, are trained using a mean squared error regression objective to match bootstrapped target values. However, scaling value-based RL methods that use regression to large networks, such as high-capacity Transformers, has proven challenging. This difficulty is in stark contrast to supervised learning: by leveraging a cross-entropy classification loss, supervised methods have scaled reliably to massive networks. Observing this discrepancy, in this paper, we investigate wheth"},"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":"2403.03950","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-03-06T18:55:47Z","cross_cats_sorted":["cs.AI","stat.ML"],"title_canon_sha256":"a88da0c353e77a2275c00a4cc1c7c7c7c1db1bca18e72dae5539d8760949cca9","abstract_canon_sha256":"e50147a7ce73aae50d8a66ca3b6455e121ecab9f281ceca0f81383aae8de3085"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:53:03.970348Z","signature_b64":"iqSguVJSNPME8rFzkUqtVv5J0HWJQnkZZN4eyHa5ljuuMxAKItFMExx58HDkRssIY3uKfM9xeJITWubeEHByDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bbd9b90c8450a16c73895b61f4db5ea74eb659870166f9b4124f36bfd778de4b","last_reissued_at":"2026-07-05T07:53:03.969870Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:53:03.969870Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Stop Regressing: Training Value Functions via Classification for Scalable Deep RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Adrien Ali Ta\\\"iga, Aleksandra Faust, Alex Irpan, Aviral Kumar, Jesse Farebrother, Jordi Orbay, Pablo Samuel Castro, Quan Vuong, Rishabh Agarwal, Sergey Levine, Ted Xiao, Yevgen Chebotar","submitted_at":"2024-03-06T18:55:47Z","abstract_excerpt":"Value functions are a central component of deep reinforcement learning (RL). These functions, parameterized by neural networks, are trained using a mean squared error regression objective to match bootstrapped target values. However, scaling value-based RL methods that use regression to large networks, such as high-capacity Transformers, has proven challenging. This difficulty is in stark contrast to supervised learning: by leveraging a cross-entropy classification loss, supervised methods have scaled reliably to massive networks. Observing this discrepancy, in this paper, we investigate wheth"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.03950","kind":"arxiv","version":1},"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/2403.03950/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":"2403.03950","created_at":"2026-07-05T07:53:03.969927+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.03950v1","created_at":"2026-07-05T07:53:03.969927+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.03950","created_at":"2026-07-05T07:53:03.969927+00:00"},{"alias_kind":"pith_short_12","alias_value":"XPM3SDEEKCQW","created_at":"2026-07-05T07:53:03.969927+00:00"},{"alias_kind":"pith_short_16","alias_value":"XPM3SDEEKCQWY44J","created_at":"2026-07-05T07:53:03.969927+00:00"},{"alias_kind":"pith_short_8","alias_value":"XPM3SDEE","created_at":"2026-07-05T07:53:03.969927+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":16,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24742","citing_title":"World Value Models for Robotic Manipulation","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2606.23348","citing_title":"Superhuman AI for Generals.io Using Self-Play Reinforcement Learning","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20411","citing_title":"Direct Advantage Estimation for Scalable and Sample-efficient Deep Reinforcement Learning","ref_index":75,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28320","citing_title":"WARP-RM: A Warp-Augmented Relative Progress Reward Model for Data Curation","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2605.31273","citing_title":"Survival Reinforcement Learning: Toward Scalable Self-Supervised RL","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2502.20349","citing_title":"Naturalistic Computational Cognitive Science: Towards generalizable models and theories that capture the full range of natural behavior","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2510.03508","citing_title":"D2 Actor Critic: Diffusion Actor Meets Distributional Critic","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2502.20349","citing_title":"Naturalistic Computational Cognitive Science: Towards generalizable models and theories that capture the full range of natural behavior","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2409.12917","citing_title":"Training Language Models to Self-Correct via Reinforcement Learning","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2602.09580","citing_title":"SERNF: Sample-Efficient Real-World Dexterous Policy Fine-Tuning via Action-Chunked Critics and Normalizing Flows","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2603.04333","citing_title":"What Does Flow Matching Bring To TD Learning?","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2403.07815","citing_title":"Chronos: Learning the Language of Time Series","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24558","citing_title":"Hierarchical Behaviour Spaces","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18978","citing_title":"Low-Rank Adaptation for Critic Learning in Off-Policy Reinforcement Learning","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01663","citing_title":"Towards Efficient and Expressive Offline RL via Flow-Anchored Noise-conditioned Q-Learning","ref_index":59,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20381","citing_title":"Distributional Value Estimation Without Target Networks for Robust Quality-Diversity","ref_index":13,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XPM3SDEEKCQWY44JLNQ7JW26U5","json":"https://pith.science/pith/XPM3SDEEKCQWY44JLNQ7JW26U5.json","graph_json":"https://pith.science/api/pith-number/XPM3SDEEKCQWY44JLNQ7JW26U5/graph.json","events_json":"https://pith.science/api/pith-number/XPM3SDEEKCQWY44JLNQ7JW26U5/events.json","paper":"https://pith.science/paper/XPM3SDEE"},"agent_actions":{"view_html":"https://pith.science/pith/XPM3SDEEKCQWY44JLNQ7JW26U5","download_json":"https://pith.science/pith/XPM3SDEEKCQWY44JLNQ7JW26U5.json","view_paper":"https://pith.science/paper/XPM3SDEE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.03950&json=true","fetch_graph":"https://pith.science/api/pith-number/XPM3SDEEKCQWY44JLNQ7JW26U5/graph.json","fetch_events":"https://pith.science/api/pith-number/XPM3SDEEKCQWY44JLNQ7JW26U5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XPM3SDEEKCQWY44JLNQ7JW26U5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XPM3SDEEKCQWY44JLNQ7JW26U5/action/storage_attestation","attest_author":"https://pith.science/pith/XPM3SDEEKCQWY44JLNQ7JW26U5/action/author_attestation","sign_citation":"https://pith.science/pith/XPM3SDEEKCQWY44JLNQ7JW26U5/action/citation_signature","submit_replication":"https://pith.science/pith/XPM3SDEEKCQWY44JLNQ7JW26U5/action/replication_record"}},"created_at":"2026-07-05T07:53:03.969927+00:00","updated_at":"2026-07-05T07:53:03.969927+00:00"}