{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ZS3O5LMH2GEJU6J575TBK2Q22K","short_pith_number":"pith:ZS3O5LMH","schema_version":"1.0","canonical_sha256":"ccb6eead87d1889a793dff66156a1ad2b65fc36c431eb9cae9f90964bf31b990","source":{"kind":"arxiv","id":"2411.15247","version":3},"attestation_state":"computed","paper":{"title":"Reward Fine-Tuning Two-Step Diffusion Models via Learning Differentiable Latent-Space Surrogate Reward","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Gengdai Liu, Huixi Zhao, Yuesong Nan, Zhiwei Jia","submitted_at":"2024-11-22T08:00:20Z","abstract_excerpt":"Recent research has shown that fine-tuning diffusion models (DMs) with arbitrary rewards, including non-differentiable ones, is feasible with reinforcement learning (RL) techniques, enabling flexible model alignment. However, applying existing RL methods to step-distilled DMs is challenging for ultra-fast ($\\le2$-step) image generation. Our analysis suggests several limitations of policy-based RL methods such as PPO or DPO toward this goal. Based on the insights, we propose fine-tuning DMs with learned differentiable surrogate rewards. Our method, named LaSRO, learns surrogate reward models in"},"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":"2411.15247","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-11-22T08:00:20Z","cross_cats_sorted":[],"title_canon_sha256":"57c679ab75ae40afa5bb877e5b5de7473abe21746dc007f89dc0a45fccd45602","abstract_canon_sha256":"5a988d4837a35569a8243bb0361f4768d1921719bdd3e48b1884355d8813fdb4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:28:26.799794Z","signature_b64":"eKEoHr8DtHfUf80AIFCEZW4Mg5Tejs4igffb3cpxqHwhIn7IyrsKewn1diqpiOzn3f/zQgoa80wKxLvkTuFiDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ccb6eead87d1889a793dff66156a1ad2b65fc36c431eb9cae9f90964bf31b990","last_reissued_at":"2026-07-05T10:28:26.799288Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:28:26.799288Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Reward Fine-Tuning Two-Step Diffusion Models via Learning Differentiable Latent-Space Surrogate Reward","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Gengdai Liu, Huixi Zhao, Yuesong Nan, Zhiwei Jia","submitted_at":"2024-11-22T08:00:20Z","abstract_excerpt":"Recent research has shown that fine-tuning diffusion models (DMs) with arbitrary rewards, including non-differentiable ones, is feasible with reinforcement learning (RL) techniques, enabling flexible model alignment. However, applying existing RL methods to step-distilled DMs is challenging for ultra-fast ($\\le2$-step) image generation. Our analysis suggests several limitations of policy-based RL methods such as PPO or DPO toward this goal. Based on the insights, we propose fine-tuning DMs with learned differentiable surrogate rewards. Our method, named LaSRO, learns surrogate reward models in"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2411.15247","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/2411.15247/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":"2411.15247","created_at":"2026-07-05T10:28:26.799368+00:00"},{"alias_kind":"arxiv_version","alias_value":"2411.15247v3","created_at":"2026-07-05T10:28:26.799368+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2411.15247","created_at":"2026-07-05T10:28:26.799368+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZS3O5LMH2GEJ","created_at":"2026-07-05T10:28:26.799368+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZS3O5LMH2GEJU6J5","created_at":"2026-07-05T10:28:26.799368+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZS3O5LMH","created_at":"2026-07-05T10:28:26.799368+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07693","citing_title":"Selective Timestep Weighting and Advantage-Based Replay for Sample-Efficient Diffusion RLHF","ref_index":17,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZS3O5LMH2GEJU6J575TBK2Q22K","json":"https://pith.science/pith/ZS3O5LMH2GEJU6J575TBK2Q22K.json","graph_json":"https://pith.science/api/pith-number/ZS3O5LMH2GEJU6J575TBK2Q22K/graph.json","events_json":"https://pith.science/api/pith-number/ZS3O5LMH2GEJU6J575TBK2Q22K/events.json","paper":"https://pith.science/paper/ZS3O5LMH"},"agent_actions":{"view_html":"https://pith.science/pith/ZS3O5LMH2GEJU6J575TBK2Q22K","download_json":"https://pith.science/pith/ZS3O5LMH2GEJU6J575TBK2Q22K.json","view_paper":"https://pith.science/paper/ZS3O5LMH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2411.15247&json=true","fetch_graph":"https://pith.science/api/pith-number/ZS3O5LMH2GEJU6J575TBK2Q22K/graph.json","fetch_events":"https://pith.science/api/pith-number/ZS3O5LMH2GEJU6J575TBK2Q22K/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZS3O5LMH2GEJU6J575TBK2Q22K/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZS3O5LMH2GEJU6J575TBK2Q22K/action/storage_attestation","attest_author":"https://pith.science/pith/ZS3O5LMH2GEJU6J575TBK2Q22K/action/author_attestation","sign_citation":"https://pith.science/pith/ZS3O5LMH2GEJU6J575TBK2Q22K/action/citation_signature","submit_replication":"https://pith.science/pith/ZS3O5LMH2GEJU6J575TBK2Q22K/action/replication_record"}},"created_at":"2026-07-05T10:28:26.799368+00:00","updated_at":"2026-07-05T10:28:26.799368+00:00"}