{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:Y3RQ5AOK6YETPMWVGSBBUEVQ7F","short_pith_number":"pith:Y3RQ5AOK","schema_version":"1.0","canonical_sha256":"c6e30e81caf60937b2d534821a12b0f95cad06d96d1da029a1d8d27679f6d268","source":{"kind":"arxiv","id":"2505.19196","version":1},"attestation_state":"computed","paper":{"title":"Step-level Reward for Free in RL-based T2I Diffusion Model Fine-tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Wei Wei, Xiaoye Qu, Xinyao Liao, Yu Cheng","submitted_at":"2025-05-25T15:43:54Z","abstract_excerpt":"Recent advances in text-to-image (T2I) diffusion model fine-tuning leverage reinforcement learning (RL) to align generated images with learnable reward functions. The existing approaches reformulate denoising as a Markov decision process for RL-driven optimization. However, they suffer from reward sparsity, receiving only a single delayed reward per generated trajectory. This flaw hinders precise step-level attribution of denoising actions, undermines training efficiency. To address this, we propose a simple yet effective credit assignment framework that dynamically distributes dense rewards a"},"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":"2505.19196","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-05-25T15:43:54Z","cross_cats_sorted":[],"title_canon_sha256":"7f958bd0ec4c97199ca320686c7eee8732686e63990534b06a81734213eb53a4","abstract_canon_sha256":"9672432cc60cea11c32325c56932a3f40bd47fc8364fc10c653035dc85cbbfb4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:09:26.319588Z","signature_b64":"mKqM481fSwxc7X1v0cJmKRbVWoZyM6+KOk9Reb0lmBtWNhj3LebD1xiQnJiw2F1624xtvt5pYjCmWjN3ZUCyBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c6e30e81caf60937b2d534821a12b0f95cad06d96d1da029a1d8d27679f6d268","last_reissued_at":"2026-07-05T11:09:26.318951Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:09:26.318951Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Step-level Reward for Free in RL-based T2I Diffusion Model Fine-tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Wei Wei, Xiaoye Qu, Xinyao Liao, Yu Cheng","submitted_at":"2025-05-25T15:43:54Z","abstract_excerpt":"Recent advances in text-to-image (T2I) diffusion model fine-tuning leverage reinforcement learning (RL) to align generated images with learnable reward functions. The existing approaches reformulate denoising as a Markov decision process for RL-driven optimization. However, they suffer from reward sparsity, receiving only a single delayed reward per generated trajectory. This flaw hinders precise step-level attribution of denoising actions, undermines training efficiency. To address this, we propose a simple yet effective credit assignment framework that dynamically distributes dense rewards a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.19196","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/2505.19196/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":"2505.19196","created_at":"2026-07-05T11:09:26.319030+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.19196v1","created_at":"2026-07-05T11:09:26.319030+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.19196","created_at":"2026-07-05T11:09:26.319030+00:00"},{"alias_kind":"pith_short_12","alias_value":"Y3RQ5AOK6YET","created_at":"2026-07-05T11:09:26.319030+00:00"},{"alias_kind":"pith_short_16","alias_value":"Y3RQ5AOK6YETPMWV","created_at":"2026-07-05T11:09:26.319030+00:00"},{"alias_kind":"pith_short_8","alias_value":"Y3RQ5AOK","created_at":"2026-07-05T11:09:26.319030+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.30317","citing_title":"VPG: Visual Prefix Guidance for Autoregressive Image and Video Generation","ref_index":27,"is_internal_anchor":true},{"citing_arxiv_id":"2604.15311","citing_title":"LeapAlign: Post-Training Flow Matching Models at Any Generation Step by Building Two-Step Trajectories","ref_index":26,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Y3RQ5AOK6YETPMWVGSBBUEVQ7F","json":"https://pith.science/pith/Y3RQ5AOK6YETPMWVGSBBUEVQ7F.json","graph_json":"https://pith.science/api/pith-number/Y3RQ5AOK6YETPMWVGSBBUEVQ7F/graph.json","events_json":"https://pith.science/api/pith-number/Y3RQ5AOK6YETPMWVGSBBUEVQ7F/events.json","paper":"https://pith.science/paper/Y3RQ5AOK"},"agent_actions":{"view_html":"https://pith.science/pith/Y3RQ5AOK6YETPMWVGSBBUEVQ7F","download_json":"https://pith.science/pith/Y3RQ5AOK6YETPMWVGSBBUEVQ7F.json","view_paper":"https://pith.science/paper/Y3RQ5AOK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.19196&json=true","fetch_graph":"https://pith.science/api/pith-number/Y3RQ5AOK6YETPMWVGSBBUEVQ7F/graph.json","fetch_events":"https://pith.science/api/pith-number/Y3RQ5AOK6YETPMWVGSBBUEVQ7F/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Y3RQ5AOK6YETPMWVGSBBUEVQ7F/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Y3RQ5AOK6YETPMWVGSBBUEVQ7F/action/storage_attestation","attest_author":"https://pith.science/pith/Y3RQ5AOK6YETPMWVGSBBUEVQ7F/action/author_attestation","sign_citation":"https://pith.science/pith/Y3RQ5AOK6YETPMWVGSBBUEVQ7F/action/citation_signature","submit_replication":"https://pith.science/pith/Y3RQ5AOK6YETPMWVGSBBUEVQ7F/action/replication_record"}},"created_at":"2026-07-05T11:09:26.319030+00:00","updated_at":"2026-07-05T11:09:26.319030+00:00"}