{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:BM7BAZZO2IKTG4QGLYJDLPJW5Q","short_pith_number":"pith:BM7BAZZO","schema_version":"1.0","canonical_sha256":"0b3e10672ed2153372065e1235bd36ec273ee2c2ad6ac8fd199d6590531605df","source":{"kind":"arxiv","id":"2502.01819","version":3},"attestation_state":"computed","paper":{"title":"Score as Action: Fine-Tuning Diffusion Generative Models by Continuous-time Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","math.OC"],"primary_cat":"cs.LG","authors_text":"David D. Yao, Hanyang Zhao, Haoxian Chen, Ji Zhang, Wenpin Tang","submitted_at":"2025-02-03T20:50:05Z","abstract_excerpt":"Reinforcement learning from human feedback (RLHF), which aligns a diffusion model with input prompt, has become a crucial step in building reliable generative AI models. Most works in this area use a discrete-time formulation, which is prone to induced discretization errors, and often not applicable to models with higher-order/black-box solvers. The objective of this study is to develop a disciplined approach to fine-tune diffusion models using continuous-time RL, formulated as a stochastic control problem with a reward function that aligns the end result (terminal state) with input prompt. Th"},"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":"2502.01819","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-03T20:50:05Z","cross_cats_sorted":["cs.AI","math.OC"],"title_canon_sha256":"8936995aab6ab488306b4d6cbb56b66e7ea9629f3f31afeaa917681d5914f480","abstract_canon_sha256":"b42f6deec2f9c94828db59afdee3eb81341b708410efe356515a021933c9b75d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:57:20.019479Z","signature_b64":"rxJdNL55oq0LfyKQeK9wuOrIpHjFdwiN+EK3fAp1PoBZ606/axWiqmqXzPJEuYHMZeCty46CE/N3N9gvIolpBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0b3e10672ed2153372065e1235bd36ec273ee2c2ad6ac8fd199d6590531605df","last_reissued_at":"2026-07-05T11:57:20.018981Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:57:20.018981Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Score as Action: Fine-Tuning Diffusion Generative Models by Continuous-time Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","math.OC"],"primary_cat":"cs.LG","authors_text":"David D. Yao, Hanyang Zhao, Haoxian Chen, Ji Zhang, Wenpin Tang","submitted_at":"2025-02-03T20:50:05Z","abstract_excerpt":"Reinforcement learning from human feedback (RLHF), which aligns a diffusion model with input prompt, has become a crucial step in building reliable generative AI models. Most works in this area use a discrete-time formulation, which is prone to induced discretization errors, and often not applicable to models with higher-order/black-box solvers. The objective of this study is to develop a disciplined approach to fine-tune diffusion models using continuous-time RL, formulated as a stochastic control problem with a reward function that aligns the end result (terminal state) with input prompt. Th"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.01819","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":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2502.01819/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":"2502.01819","created_at":"2026-07-05T11:57:20.019037+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.01819v3","created_at":"2026-07-05T11:57:20.019037+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.01819","created_at":"2026-07-05T11:57:20.019037+00:00"},{"alias_kind":"pith_short_12","alias_value":"BM7BAZZO2IKT","created_at":"2026-07-05T11:57:20.019037+00:00"},{"alias_kind":"pith_short_16","alias_value":"BM7BAZZO2IKTG4QG","created_at":"2026-07-05T11:57:20.019037+00:00"},{"alias_kind":"pith_short_8","alias_value":"BM7BAZZO","created_at":"2026-07-05T11:57:20.019037+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.15803","citing_title":"Embedding-perturbed Exploration Preference Optimization for Flow Models","ref_index":97,"is_internal_anchor":false},{"citing_arxiv_id":"2603.00918","citing_title":"Improving Text-to-Image Generation with Intrinsic Self-Confidence Rewards","ref_index":89,"is_internal_anchor":false},{"citing_arxiv_id":"2505.05470","citing_title":"Flow-GRPO: Training Flow Matching Models via Online RL","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05530","citing_title":"Energy Generative Modeling: A Lyapunov-based Energy Matching Perspective","ref_index":41,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BM7BAZZO2IKTG4QGLYJDLPJW5Q","json":"https://pith.science/pith/BM7BAZZO2IKTG4QGLYJDLPJW5Q.json","graph_json":"https://pith.science/api/pith-number/BM7BAZZO2IKTG4QGLYJDLPJW5Q/graph.json","events_json":"https://pith.science/api/pith-number/BM7BAZZO2IKTG4QGLYJDLPJW5Q/events.json","paper":"https://pith.science/paper/BM7BAZZO"},"agent_actions":{"view_html":"https://pith.science/pith/BM7BAZZO2IKTG4QGLYJDLPJW5Q","download_json":"https://pith.science/pith/BM7BAZZO2IKTG4QGLYJDLPJW5Q.json","view_paper":"https://pith.science/paper/BM7BAZZO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.01819&json=true","fetch_graph":"https://pith.science/api/pith-number/BM7BAZZO2IKTG4QGLYJDLPJW5Q/graph.json","fetch_events":"https://pith.science/api/pith-number/BM7BAZZO2IKTG4QGLYJDLPJW5Q/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BM7BAZZO2IKTG4QGLYJDLPJW5Q/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BM7BAZZO2IKTG4QGLYJDLPJW5Q/action/storage_attestation","attest_author":"https://pith.science/pith/BM7BAZZO2IKTG4QGLYJDLPJW5Q/action/author_attestation","sign_citation":"https://pith.science/pith/BM7BAZZO2IKTG4QGLYJDLPJW5Q/action/citation_signature","submit_replication":"https://pith.science/pith/BM7BAZZO2IKTG4QGLYJDLPJW5Q/action/replication_record"}},"created_at":"2026-07-05T11:57:20.019037+00:00","updated_at":"2026-07-05T11:57:20.019037+00:00"}