{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:EYFDE73TE3V2WFEFSOGZZ3WQ4M","short_pith_number":"pith:EYFDE73T","schema_version":"1.0","canonical_sha256":"260a327f7326ebab1485938d9ceed0e32b0256bef508462b3606c79affd9835f","source":{"kind":"arxiv","id":"2502.01667","version":1},"attestation_state":"computed","paper":{"title":"Refining Alignment Framework for Diffusion Models with Intermediate-Step Preference Ranking","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Dongrui Liu, Jie Ren, Qi Tian, Xiaopeng Zhang, Yuhang Zhang","submitted_at":"2025-02-01T16:08:43Z","abstract_excerpt":"Direct preference optimization (DPO) has shown success in aligning diffusion models with human preference. Previous approaches typically assume a consistent preference label between final generations and noisy samples at intermediate steps, and directly apply DPO to these noisy samples for fine-tuning. However, we theoretically identify inherent issues in this assumption and its impacts on the effectiveness of preference alignment. We first demonstrate the inherent issues from two perspectives: gradient direction and preference order, and then propose a Tailored Preference Optimization (Tailor"},"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":"2502.01667","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-01T16:08:43Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"16494a6684644de5e834de5c38c7dd6a7f39d3d07aade5e685e2c5ebeac45535","abstract_canon_sha256":"aaa01598d7dab8d09797dba529eeec987cf786a5ed6b52d7f3c6fb9823714772"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:09:17.323999Z","signature_b64":"SemSoQVeckQsxUfYVulK9cSKaxiWU6ooo7tfQ7cKr4EEX/fof8vmOpF+TCght8qb6brI0KpjFpUOCUb2tb0EDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"260a327f7326ebab1485938d9ceed0e32b0256bef508462b3606c79affd9835f","last_reissued_at":"2026-07-05T10:09:17.323586Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:09:17.323586Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Refining Alignment Framework for Diffusion Models with Intermediate-Step Preference Ranking","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Dongrui Liu, Jie Ren, Qi Tian, Xiaopeng Zhang, Yuhang Zhang","submitted_at":"2025-02-01T16:08:43Z","abstract_excerpt":"Direct preference optimization (DPO) has shown success in aligning diffusion models with human preference. Previous approaches typically assume a consistent preference label between final generations and noisy samples at intermediate steps, and directly apply DPO to these noisy samples for fine-tuning. However, we theoretically identify inherent issues in this assumption and its impacts on the effectiveness of preference alignment. We first demonstrate the inherent issues from two perspectives: gradient direction and preference order, and then propose a Tailored Preference Optimization (Tailor"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.01667","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/2502.01667/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":"2502.01667","created_at":"2026-07-05T10:09:17.323645+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.01667v1","created_at":"2026-07-05T10:09:17.323645+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.01667","created_at":"2026-07-05T10:09:17.323645+00:00"},{"alias_kind":"pith_short_12","alias_value":"EYFDE73TE3V2","created_at":"2026-07-05T10:09:17.323645+00:00"},{"alias_kind":"pith_short_16","alias_value":"EYFDE73TE3V2WFEF","created_at":"2026-07-05T10:09:17.323645+00:00"},{"alias_kind":"pith_short_8","alias_value":"EYFDE73T","created_at":"2026-07-05T10:09:17.323645+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.09433","citing_title":"Offline Preference Optimization for Rectified Flow with Noise-Tracked Pairs","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04653","citing_title":"Threshold-Guided Optimization for Visual Generative Models","ref_index":47,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EYFDE73TE3V2WFEFSOGZZ3WQ4M","json":"https://pith.science/pith/EYFDE73TE3V2WFEFSOGZZ3WQ4M.json","graph_json":"https://pith.science/api/pith-number/EYFDE73TE3V2WFEFSOGZZ3WQ4M/graph.json","events_json":"https://pith.science/api/pith-number/EYFDE73TE3V2WFEFSOGZZ3WQ4M/events.json","paper":"https://pith.science/paper/EYFDE73T"},"agent_actions":{"view_html":"https://pith.science/pith/EYFDE73TE3V2WFEFSOGZZ3WQ4M","download_json":"https://pith.science/pith/EYFDE73TE3V2WFEFSOGZZ3WQ4M.json","view_paper":"https://pith.science/paper/EYFDE73T","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.01667&json=true","fetch_graph":"https://pith.science/api/pith-number/EYFDE73TE3V2WFEFSOGZZ3WQ4M/graph.json","fetch_events":"https://pith.science/api/pith-number/EYFDE73TE3V2WFEFSOGZZ3WQ4M/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EYFDE73TE3V2WFEFSOGZZ3WQ4M/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EYFDE73TE3V2WFEFSOGZZ3WQ4M/action/storage_attestation","attest_author":"https://pith.science/pith/EYFDE73TE3V2WFEFSOGZZ3WQ4M/action/author_attestation","sign_citation":"https://pith.science/pith/EYFDE73TE3V2WFEFSOGZZ3WQ4M/action/citation_signature","submit_replication":"https://pith.science/pith/EYFDE73TE3V2WFEFSOGZZ3WQ4M/action/replication_record"}},"created_at":"2026-07-05T10:09:17.323645+00:00","updated_at":"2026-07-05T10:09:17.323645+00:00"}