{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:RQ4AQ6KGRT4TBOY3HURYNLFMP6","short_pith_number":"pith:RQ4AQ6KG","schema_version":"1.0","canonical_sha256":"8c380879468cf930bb1b3d2386acac7faf37e97ba9b8a8bc55123e59243448eb","source":{"kind":"arxiv","id":"2410.21662","version":2},"attestation_state":"computed","paper":{"title":"$f$-PO: Generalizing Preference Optimization with $f$-divergence Minimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Jiaqi Han, Mingjian Jiang, Minkai Xu, Stefano Ermon, Yuxuan Song","submitted_at":"2024-10-29T02:11:45Z","abstract_excerpt":"Preference optimization has made significant progress recently, with numerous methods developed to align language models with human preferences. This paper introduces $f$-divergence Preference Optimization ($f$-PO), a novel framework that generalizes and extends existing approaches. $f$-PO minimizes $f$-divergences between the optimized policy and the optimal policy, encompassing a broad family of alignment methods using various divergences. Our approach unifies previous algorithms like DPO and EXO, while offering new variants through different choices of $f$-divergences. We provide theoretica"},"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":"2410.21662","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-29T02:11:45Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"f03d4edda1cbddb55b3e41b15544148895cedc74278896c63a24b02b8461dd0b","abstract_canon_sha256":"6f0855a0f907c265fb2c596a60915e060084211015d1487f643d7b643b28e714"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:14:47.359858Z","signature_b64":"Im/B7AfclyKNXat8zoS5q/+p+bBwfuTMFQJKwhN/M9shVzCRu1F9eqLLaHDTP5Yw2gnKxfpGDivUF4cAxyyIBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8c380879468cf930bb1b3d2386acac7faf37e97ba9b8a8bc55123e59243448eb","last_reissued_at":"2026-07-05T10:14:47.359192Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:14:47.359192Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"$f$-PO: Generalizing Preference Optimization with $f$-divergence Minimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Jiaqi Han, Mingjian Jiang, Minkai Xu, Stefano Ermon, Yuxuan Song","submitted_at":"2024-10-29T02:11:45Z","abstract_excerpt":"Preference optimization has made significant progress recently, with numerous methods developed to align language models with human preferences. This paper introduces $f$-divergence Preference Optimization ($f$-PO), a novel framework that generalizes and extends existing approaches. $f$-PO minimizes $f$-divergences between the optimized policy and the optimal policy, encompassing a broad family of alignment methods using various divergences. Our approach unifies previous algorithms like DPO and EXO, while offering new variants through different choices of $f$-divergences. We provide theoretica"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.21662","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/2410.21662/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":"2410.21662","created_at":"2026-07-05T10:14:47.359274+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.21662v2","created_at":"2026-07-05T10:14:47.359274+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.21662","created_at":"2026-07-05T10:14:47.359274+00:00"},{"alias_kind":"pith_short_12","alias_value":"RQ4AQ6KGRT4T","created_at":"2026-07-05T10:14:47.359274+00:00"},{"alias_kind":"pith_short_16","alias_value":"RQ4AQ6KGRT4TBOY3","created_at":"2026-07-05T10:14:47.359274+00:00"},{"alias_kind":"pith_short_8","alias_value":"RQ4AQ6KG","created_at":"2026-07-05T10:14:47.359274+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2502.01237","citing_title":"The Differences Between Direct Alignment Algorithms are a Blur","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06977","citing_title":"$f$-Divergence Regularized RLHF: Two Tales of Sampling and Unified Analyses","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RQ4AQ6KGRT4TBOY3HURYNLFMP6","json":"https://pith.science/pith/RQ4AQ6KGRT4TBOY3HURYNLFMP6.json","graph_json":"https://pith.science/api/pith-number/RQ4AQ6KGRT4TBOY3HURYNLFMP6/graph.json","events_json":"https://pith.science/api/pith-number/RQ4AQ6KGRT4TBOY3HURYNLFMP6/events.json","paper":"https://pith.science/paper/RQ4AQ6KG"},"agent_actions":{"view_html":"https://pith.science/pith/RQ4AQ6KGRT4TBOY3HURYNLFMP6","download_json":"https://pith.science/pith/RQ4AQ6KGRT4TBOY3HURYNLFMP6.json","view_paper":"https://pith.science/paper/RQ4AQ6KG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.21662&json=true","fetch_graph":"https://pith.science/api/pith-number/RQ4AQ6KGRT4TBOY3HURYNLFMP6/graph.json","fetch_events":"https://pith.science/api/pith-number/RQ4AQ6KGRT4TBOY3HURYNLFMP6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RQ4AQ6KGRT4TBOY3HURYNLFMP6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RQ4AQ6KGRT4TBOY3HURYNLFMP6/action/storage_attestation","attest_author":"https://pith.science/pith/RQ4AQ6KGRT4TBOY3HURYNLFMP6/action/author_attestation","sign_citation":"https://pith.science/pith/RQ4AQ6KGRT4TBOY3HURYNLFMP6/action/citation_signature","submit_replication":"https://pith.science/pith/RQ4AQ6KGRT4TBOY3HURYNLFMP6/action/replication_record"}},"created_at":"2026-07-05T10:14:47.359274+00:00","updated_at":"2026-07-05T10:14:47.359274+00:00"}