{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:BQRKYGZPTMCHFEQURWNZ5IDSRN","short_pith_number":"pith:BQRKYGZP","schema_version":"1.0","canonical_sha256":"0c22ac1b2f9b047292148d9b9ea0728b53efed9ecf88a7da5f237960e8fe1a08","source":{"kind":"arxiv","id":"2409.08400","version":1},"attestation_state":"computed","paper":{"title":"Scores as Actions: a framework of fine-tuning diffusion models by continuous-time reinforcement learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"David D. Yao, Hanyang Zhao, Haoxian Chen, Ji Zhang, Wenpin Tang","submitted_at":"2024-09-12T21:12:21Z","abstract_excerpt":"Reinforcement Learning from human feedback (RLHF) has been shown a promising direction for aligning generative models with human intent and has also been explored in recent works for alignment of diffusion generative models. In this work, we provide a rigorous treatment by formulating the task of fine-tuning diffusion models, with reward functions learned from human feedback, as an exploratory continuous-time stochastic control problem. Our key idea lies in treating the score-matching functions as controls/actions, and upon this, we develop a unified framework from a continuous-time perspectiv"},"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":"2409.08400","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-12T21:12:21Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"47c507b5ce44b15c3c74d7ca1ab5f36ee5960d4d4c5a2a1d5a77a828e90ff647","abstract_canon_sha256":"55a5ca6d7dc458c08056ab7f9e661a260b16217442ef7248034d0629b593b803"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:06:30.570086Z","signature_b64":"O+EMsrFMfMNJbQ6mCWoOmH0TZPKtveeFScBg97LG3gJgFjj+WnuvLHaLNyX3BgdLKk6KqA1P2TckzGKH78/pDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0c22ac1b2f9b047292148d9b9ea0728b53efed9ecf88a7da5f237960e8fe1a08","last_reissued_at":"2026-07-05T09:06:30.569571Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:06:30.569571Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Scores as Actions: a framework of fine-tuning diffusion models by continuous-time reinforcement learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"David D. Yao, Hanyang Zhao, Haoxian Chen, Ji Zhang, Wenpin Tang","submitted_at":"2024-09-12T21:12:21Z","abstract_excerpt":"Reinforcement Learning from human feedback (RLHF) has been shown a promising direction for aligning generative models with human intent and has also been explored in recent works for alignment of diffusion generative models. In this work, we provide a rigorous treatment by formulating the task of fine-tuning diffusion models, with reward functions learned from human feedback, as an exploratory continuous-time stochastic control problem. Our key idea lies in treating the score-matching functions as controls/actions, and upon this, we develop a unified framework from a continuous-time perspectiv"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.08400","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/2409.08400/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":"2409.08400","created_at":"2026-07-05T09:06:30.569639+00:00"},{"alias_kind":"arxiv_version","alias_value":"2409.08400v1","created_at":"2026-07-05T09:06:30.569639+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.08400","created_at":"2026-07-05T09:06:30.569639+00:00"},{"alias_kind":"pith_short_12","alias_value":"BQRKYGZPTMCH","created_at":"2026-07-05T09:06:30.569639+00:00"},{"alias_kind":"pith_short_16","alias_value":"BQRKYGZPTMCHFEQU","created_at":"2026-07-05T09:06:30.569639+00:00"},{"alias_kind":"pith_short_8","alias_value":"BQRKYGZP","created_at":"2026-07-05T09:06:30.569639+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.02137","citing_title":"ART for Diffusion Sampling: Continuous-Time Control and Actor-Critic Learning","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2601.18681","citing_title":"ART for Diffusion Sampling: A Reinforcement Learning Approach to Timestep Schedule","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2602.22507","citing_title":"Space Syntax-guided Post-training for Residential Floor Plan Generation","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06583","citing_title":"Improved techniques for fine-tuning flow models via adjoint matching: a deterministic control pipeline","ref_index":53,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BQRKYGZPTMCHFEQURWNZ5IDSRN","json":"https://pith.science/pith/BQRKYGZPTMCHFEQURWNZ5IDSRN.json","graph_json":"https://pith.science/api/pith-number/BQRKYGZPTMCHFEQURWNZ5IDSRN/graph.json","events_json":"https://pith.science/api/pith-number/BQRKYGZPTMCHFEQURWNZ5IDSRN/events.json","paper":"https://pith.science/paper/BQRKYGZP"},"agent_actions":{"view_html":"https://pith.science/pith/BQRKYGZPTMCHFEQURWNZ5IDSRN","download_json":"https://pith.science/pith/BQRKYGZPTMCHFEQURWNZ5IDSRN.json","view_paper":"https://pith.science/paper/BQRKYGZP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2409.08400&json=true","fetch_graph":"https://pith.science/api/pith-number/BQRKYGZPTMCHFEQURWNZ5IDSRN/graph.json","fetch_events":"https://pith.science/api/pith-number/BQRKYGZPTMCHFEQURWNZ5IDSRN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BQRKYGZPTMCHFEQURWNZ5IDSRN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BQRKYGZPTMCHFEQURWNZ5IDSRN/action/storage_attestation","attest_author":"https://pith.science/pith/BQRKYGZPTMCHFEQURWNZ5IDSRN/action/author_attestation","sign_citation":"https://pith.science/pith/BQRKYGZPTMCHFEQURWNZ5IDSRN/action/citation_signature","submit_replication":"https://pith.science/pith/BQRKYGZPTMCHFEQURWNZ5IDSRN/action/replication_record"}},"created_at":"2026-07-05T09:06:30.569639+00:00","updated_at":"2026-07-05T09:06:30.569639+00:00"}