{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:XMD5YC5PVLLJ2SQNWOVCZ4TJFU","short_pith_number":"pith:XMD5YC5P","schema_version":"1.0","canonical_sha256":"bb07dc0bafaad69d4a0db3aa2cf2692d1c92cc5f32433d278f7b75d9dd4549ef","source":{"kind":"arxiv","id":"2602.04663","version":2},"attestation_state":"computed","paper":{"title":"Rethinking the Design Space of Reinforcement Learning for Diffusion Models: On the Importance of Likelihood Estimation Beyond Loss Design","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Bo Yuan, Jaemoo Choi, Jinbin Bai, Molei Tao, Petr Molodyk, Wei Guo, Yi Xin, Yongxin Chen, Yuchen Zhu","submitted_at":"2026-02-04T15:36:42Z","abstract_excerpt":"Reinforcement learning has been widely applied to diffusion and flow models for visual tasks such as text-to-image generation. However, these tasks remain challenging because diffusion models have intractable likelihoods, which creates a barrier for directly applying popular policy-gradient type methods. Existing approaches primarily focus on crafting new objectives built on already heavily engineered LLM objectives, using ad hoc estimators for likelihood, without a thorough investigation into how such estimation affects overall algorithmic performance. In this work, we provide a systematic an"},"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":"2602.04663","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-04T15:36:42Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"cb2c90a5aeb9c3df9aa52ad73b058f979ef8fb140fbd0f9281cc6c91ebe82503","abstract_canon_sha256":"7ba8f3cfc2c8f70596a1c781f0239d67472462b78cfbc0027c0bf8ca733e2104"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T01:05:08.135084Z","signature_b64":"/2OZb/OAMwaDpgYsAXLsWHs4AlxJMfoRw8wS6OXM09Iea2e1NLq9QKuQ9Lbub8uOD1aFrQezPxJ6ZmNOuMn8Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bb07dc0bafaad69d4a0db3aa2cf2692d1c92cc5f32433d278f7b75d9dd4549ef","last_reissued_at":"2026-05-20T01:05:08.134258Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T01:05:08.134258Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Rethinking the Design Space of Reinforcement Learning for Diffusion Models: On the Importance of Likelihood Estimation Beyond Loss Design","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Bo Yuan, Jaemoo Choi, Jinbin Bai, Molei Tao, Petr Molodyk, Wei Guo, Yi Xin, Yongxin Chen, Yuchen Zhu","submitted_at":"2026-02-04T15:36:42Z","abstract_excerpt":"Reinforcement learning has been widely applied to diffusion and flow models for visual tasks such as text-to-image generation. However, these tasks remain challenging because diffusion models have intractable likelihoods, which creates a barrier for directly applying popular policy-gradient type methods. Existing approaches primarily focus on crafting new objectives built on already heavily engineered LLM objectives, using ad hoc estimators for likelihood, without a thorough investigation into how such estimation affects overall algorithmic performance. In this work, we provide a systematic an"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.04663","kind":"arxiv","version":2},"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/2602.04663/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":"2602.04663","created_at":"2026-05-20T01:05:08.134374+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.04663v2","created_at":"2026-05-20T01:05:08.134374+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.04663","created_at":"2026-05-20T01:05:08.134374+00:00"},{"alias_kind":"pith_short_12","alias_value":"XMD5YC5PVLLJ","created_at":"2026-05-20T01:05:08.134374+00:00"},{"alias_kind":"pith_short_16","alias_value":"XMD5YC5PVLLJ2SQN","created_at":"2026-05-20T01:05:08.134374+00:00"},{"alias_kind":"pith_short_8","alias_value":"XMD5YC5P","created_at":"2026-05-20T01:05:08.134374+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":5,"sample":[{"citing_arxiv_id":"2605.10759","citing_title":"Reinforce Adjoint Matching: Scaling RL Post-Training of Diffusion and Flow-Matching Models","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10759","citing_title":"Reinforce Adjoint Matching: Scaling RL Post-Training of Diffusion and Flow-Matching Models","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2604.06916","citing_title":"FP4 Explore, BF16 Train: Diffusion Reinforcement Learning via Efficient Rollout Scaling","ref_index":48,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07907","citing_title":"Consistency Regularised Gradient Flows for Inverse Problems","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2604.17415","citing_title":"Reward Score Matching: Unifying Reward-based Fine-tuning for Flow and Diffusion Models","ref_index":4,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XMD5YC5PVLLJ2SQNWOVCZ4TJFU","json":"https://pith.science/pith/XMD5YC5PVLLJ2SQNWOVCZ4TJFU.json","graph_json":"https://pith.science/api/pith-number/XMD5YC5PVLLJ2SQNWOVCZ4TJFU/graph.json","events_json":"https://pith.science/api/pith-number/XMD5YC5PVLLJ2SQNWOVCZ4TJFU/events.json","paper":"https://pith.science/paper/XMD5YC5P"},"agent_actions":{"view_html":"https://pith.science/pith/XMD5YC5PVLLJ2SQNWOVCZ4TJFU","download_json":"https://pith.science/pith/XMD5YC5PVLLJ2SQNWOVCZ4TJFU.json","view_paper":"https://pith.science/paper/XMD5YC5P","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.04663&json=true","fetch_graph":"https://pith.science/api/pith-number/XMD5YC5PVLLJ2SQNWOVCZ4TJFU/graph.json","fetch_events":"https://pith.science/api/pith-number/XMD5YC5PVLLJ2SQNWOVCZ4TJFU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XMD5YC5PVLLJ2SQNWOVCZ4TJFU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XMD5YC5PVLLJ2SQNWOVCZ4TJFU/action/storage_attestation","attest_author":"https://pith.science/pith/XMD5YC5PVLLJ2SQNWOVCZ4TJFU/action/author_attestation","sign_citation":"https://pith.science/pith/XMD5YC5PVLLJ2SQNWOVCZ4TJFU/action/citation_signature","submit_replication":"https://pith.science/pith/XMD5YC5PVLLJ2SQNWOVCZ4TJFU/action/replication_record"}},"created_at":"2026-05-20T01:05:08.134374+00:00","updated_at":"2026-05-20T01:05:08.134374+00:00"}