{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2016:ASEWOIVFLCO6AKI2RI6GQ4XZ5I","short_pith_number":"pith:ASEWOIVF","schema_version":"1.0","canonical_sha256":"04896722a5589de0291a8a3c6872f9ea0456e3ac4ef0b08ea48d71dee7bc943b","source":{"kind":"arxiv","id":"1611.01626","version":3},"attestation_state":"computed","paper":{"title":"Combining policy gradient and Q-learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Brendan O'Donoghue, Koray Kavukcuoglu, Remi Munos, Volodymyr Mnih","submitted_at":"2016-11-05T10:49:37Z","abstract_excerpt":"Policy gradient is an efficient technique for improving a policy in a reinforcement learning setting. However, vanilla online variants are on-policy only and not able to take advantage of off-policy data. In this paper we describe a new technique that combines policy gradient with off-policy Q-learning, drawing experience from a replay buffer. This is motivated by making a connection between the fixed points of the regularized policy gradient algorithm and the Q-values. This connection allows us to estimate the Q-values from the action preferences of the policy, to which we apply Q-learning up"},"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":"1611.01626","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2016-11-05T10:49:37Z","cross_cats_sorted":["cs.AI","math.OC","stat.ML"],"title_canon_sha256":"53d1c180101c3f87b336ff40836f67ecb4444f7c53dd049173e1159817166549","abstract_canon_sha256":"bb8570b77a429a0fdd6770c824708c31a022807145640334d36ea6bc4b0e25a2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T00:46:51.552853Z","signature_b64":"5damb4QIINf4bNMjzTGRMrIDlerRFhRCGZWLr3rMv4dQAtloBAyV/b/QEwq81tGRh8oAXujldrzBjpkebb8iAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"04896722a5589de0291a8a3c6872f9ea0456e3ac4ef0b08ea48d71dee7bc943b","last_reissued_at":"2026-05-18T00:46:51.552236Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T00:46:51.552236Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Combining policy gradient and Q-learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Brendan O'Donoghue, Koray Kavukcuoglu, Remi Munos, Volodymyr Mnih","submitted_at":"2016-11-05T10:49:37Z","abstract_excerpt":"Policy gradient is an efficient technique for improving a policy in a reinforcement learning setting. However, vanilla online variants are on-policy only and not able to take advantage of off-policy data. In this paper we describe a new technique that combines policy gradient with off-policy Q-learning, drawing experience from a replay buffer. This is motivated by making a connection between the fixed points of the regularized policy gradient algorithm and the Q-values. This connection allows us to estimate the Q-values from the action preferences of the policy, to which we apply Q-learning up"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1611.01626","kind":"arxiv","version":3},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"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":"1611.01626","created_at":"2026-05-18T00:46:51.552335+00:00"},{"alias_kind":"arxiv_version","alias_value":"1611.01626v3","created_at":"2026-05-18T00:46:51.552335+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1611.01626","created_at":"2026-05-18T00:46:51.552335+00:00"},{"alias_kind":"pith_short_12","alias_value":"ASEWOIVFLCO6","created_at":"2026-05-18T12:30:07.202191+00:00"},{"alias_kind":"pith_short_16","alias_value":"ASEWOIVFLCO6AKI2","created_at":"2026-05-18T12:30:07.202191+00:00"},{"alias_kind":"pith_short_8","alias_value":"ASEWOIVF","created_at":"2026-05-18T12:30:07.202191+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2509.20265","citing_title":"Failure Modes of Maximum Entropy RLHF","ref_index":36,"is_internal_anchor":true},{"citing_arxiv_id":"1801.01290","citing_title":"Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13780","citing_title":"Soft $Q(\\lambda)$: A multi-step off-policy method for entropy regularised reinforcement learning using eligibility traces","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ASEWOIVFLCO6AKI2RI6GQ4XZ5I","json":"https://pith.science/pith/ASEWOIVFLCO6AKI2RI6GQ4XZ5I.json","graph_json":"https://pith.science/api/pith-number/ASEWOIVFLCO6AKI2RI6GQ4XZ5I/graph.json","events_json":"https://pith.science/api/pith-number/ASEWOIVFLCO6AKI2RI6GQ4XZ5I/events.json","paper":"https://pith.science/paper/ASEWOIVF"},"agent_actions":{"view_html":"https://pith.science/pith/ASEWOIVFLCO6AKI2RI6GQ4XZ5I","download_json":"https://pith.science/pith/ASEWOIVFLCO6AKI2RI6GQ4XZ5I.json","view_paper":"https://pith.science/paper/ASEWOIVF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1611.01626&json=true","fetch_graph":"https://pith.science/api/pith-number/ASEWOIVFLCO6AKI2RI6GQ4XZ5I/graph.json","fetch_events":"https://pith.science/api/pith-number/ASEWOIVFLCO6AKI2RI6GQ4XZ5I/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ASEWOIVFLCO6AKI2RI6GQ4XZ5I/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ASEWOIVFLCO6AKI2RI6GQ4XZ5I/action/storage_attestation","attest_author":"https://pith.science/pith/ASEWOIVFLCO6AKI2RI6GQ4XZ5I/action/author_attestation","sign_citation":"https://pith.science/pith/ASEWOIVFLCO6AKI2RI6GQ4XZ5I/action/citation_signature","submit_replication":"https://pith.science/pith/ASEWOIVFLCO6AKI2RI6GQ4XZ5I/action/replication_record"}},"created_at":"2026-05-18T00:46:51.552335+00:00","updated_at":"2026-05-18T00:46:51.552335+00:00"}