{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:HFSUFVQXRS24JHXNHQPXKTDWRN","short_pith_number":"pith:HFSUFVQX","schema_version":"1.0","canonical_sha256":"396542d6178cb5c49eed3c1f754c768b4e35378cde1576d466d4e65781f27191","source":{"kind":"arxiv","id":"2505.10892","version":2},"attestation_state":"computed","paper":{"title":"Multi-Objective Preference Optimization: Improving Human Alignment of Generative Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Akhil Agnihotri, Deepak Ramachandran, Rahul Jain, Zheng Wen","submitted_at":"2025-05-16T05:58:26Z","abstract_excerpt":"Post-training LLMs with RLHF and preference optimization methods (e.g., DPO, IPO) has greatly improved alignment, yet these approaches assume a single objective. In reality, humans express multiple, often conflicting objectives, such as helpfulness and harmlessness, with no natural scalarization. We study the multi-objective preference alignment problem, where a policy must balance several objectives simultaneously. We propose Multi-Objective Preference Optimization (MOPO), a constrained KL-regularized framework that maximizes a primary objective while enforcing lower bounds on secondary objec"},"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":"2505.10892","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-16T05:58:26Z","cross_cats_sorted":[],"title_canon_sha256":"cc1905f928e4d4bb331fb771ffcb6819d96a47c161bc473d41d6229e77786cf9","abstract_canon_sha256":"9fc854972dd4426d88776975c6cd4fd9bcd1ad93f9a437b4a37d7b862db726b9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-08T01:03:44.423454Z","signature_b64":"N+NAAMLAMQfdQeaFEu1vEac8z1J0pXeiNbfzBDWExtntPB35is9Y4ih9S7gqG0/qEt6RrAvMZG9vaPvu/itXCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"396542d6178cb5c49eed3c1f754c768b4e35378cde1576d466d4e65781f27191","last_reissued_at":"2026-06-08T01:03:44.422455Z","signature_status":"signed_v1","first_computed_at":"2026-06-08T01:03:44.422455Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Multi-Objective Preference Optimization: Improving Human Alignment of Generative Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Akhil Agnihotri, Deepak Ramachandran, Rahul Jain, Zheng Wen","submitted_at":"2025-05-16T05:58:26Z","abstract_excerpt":"Post-training LLMs with RLHF and preference optimization methods (e.g., DPO, IPO) has greatly improved alignment, yet these approaches assume a single objective. In reality, humans express multiple, often conflicting objectives, such as helpfulness and harmlessness, with no natural scalarization. We study the multi-objective preference alignment problem, where a policy must balance several objectives simultaneously. We propose Multi-Objective Preference Optimization (MOPO), a constrained KL-regularized framework that maximizes a primary objective while enforcing lower bounds on secondary objec"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.10892","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/2505.10892/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":"2505.10892","created_at":"2026-06-08T01:03:44.422597+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.10892v2","created_at":"2026-06-08T01:03:44.422597+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.10892","created_at":"2026-06-08T01:03:44.422597+00:00"},{"alias_kind":"pith_short_12","alias_value":"HFSUFVQXRS24","created_at":"2026-06-08T01:03:44.422597+00:00"},{"alias_kind":"pith_short_16","alias_value":"HFSUFVQXRS24JHXN","created_at":"2026-06-08T01:03:44.422597+00:00"},{"alias_kind":"pith_short_8","alias_value":"HFSUFVQX","created_at":"2026-06-08T01:03:44.422597+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2604.20685","citing_title":"MGDA-Decoupled: Geometry-Aware Multi-Objective Optimisation for DPO-based LLM Alignment","ref_index":74,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HFSUFVQXRS24JHXNHQPXKTDWRN","json":"https://pith.science/pith/HFSUFVQXRS24JHXNHQPXKTDWRN.json","graph_json":"https://pith.science/api/pith-number/HFSUFVQXRS24JHXNHQPXKTDWRN/graph.json","events_json":"https://pith.science/api/pith-number/HFSUFVQXRS24JHXNHQPXKTDWRN/events.json","paper":"https://pith.science/paper/HFSUFVQX"},"agent_actions":{"view_html":"https://pith.science/pith/HFSUFVQXRS24JHXNHQPXKTDWRN","download_json":"https://pith.science/pith/HFSUFVQXRS24JHXNHQPXKTDWRN.json","view_paper":"https://pith.science/paper/HFSUFVQX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.10892&json=true","fetch_graph":"https://pith.science/api/pith-number/HFSUFVQXRS24JHXNHQPXKTDWRN/graph.json","fetch_events":"https://pith.science/api/pith-number/HFSUFVQXRS24JHXNHQPXKTDWRN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HFSUFVQXRS24JHXNHQPXKTDWRN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HFSUFVQXRS24JHXNHQPXKTDWRN/action/storage_attestation","attest_author":"https://pith.science/pith/HFSUFVQXRS24JHXNHQPXKTDWRN/action/author_attestation","sign_citation":"https://pith.science/pith/HFSUFVQXRS24JHXNHQPXKTDWRN/action/citation_signature","submit_replication":"https://pith.science/pith/HFSUFVQXRS24JHXNHQPXKTDWRN/action/replication_record"}},"created_at":"2026-06-08T01:03:44.422597+00:00","updated_at":"2026-06-08T01:03:44.422597+00:00"}