{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:P5I3RYXPJ7EZMWSYUQRJPJK3PD","short_pith_number":"pith:P5I3RYXP","schema_version":"1.0","canonical_sha256":"7f51b8e2ef4fc9965a58a42297a55b78eea66056f5e390b90e56a9aedbf44467","source":{"kind":"arxiv","id":"2506.05260","version":1},"attestation_state":"computed","paper":{"title":"LeanPO: Lean Preference Optimization for Likelihood Alignment in Video-LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Jinfa Huang, Li Yuan, Peixi Peng, Xiaodong Wang","submitted_at":"2025-06-05T17:21:16Z","abstract_excerpt":"Most Video Large Language Models (Video-LLMs) adopt preference alignment techniques, e.g., DPO~\\citep{rafailov2024dpo}, to optimize the reward margin between a winning response ($y_w$) and a losing response ($y_l$). However, the likelihood displacement observed in DPO indicates that both $\\log \\pi_\\theta (y_w\\mid x)$ and $\\log \\pi_\\theta (y_l\\mid x) $ often decrease during training, inadvertently boosting the probabilities of non-target responses. In this paper, we systematically revisit this phenomenon from LLMs to Video-LLMs, showing that it intensifies when dealing with the redundant comple"},"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":"2506.05260","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-06-05T17:21:16Z","cross_cats_sorted":[],"title_canon_sha256":"2360c5dd49177ea61a9a32ab73a99250482a8a32761f6e4c66437befd8f0e86e","abstract_canon_sha256":"c9dbb85e25b7b0ef1ab2de1d92b80270a1d209915da9cabe3eb99441278a959a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:16:44.922018Z","signature_b64":"YKvs/MA4d04hoP6R52zZW+auCnc99S54x9f82GIPq/vI1zA9ttiPOON8o4xcI4kWO13aBn8aLCkbSrdhGgy6BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7f51b8e2ef4fc9965a58a42297a55b78eea66056f5e390b90e56a9aedbf44467","last_reissued_at":"2026-07-05T11:16:44.921532Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:16:44.921532Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LeanPO: Lean Preference Optimization for Likelihood Alignment in Video-LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Jinfa Huang, Li Yuan, Peixi Peng, Xiaodong Wang","submitted_at":"2025-06-05T17:21:16Z","abstract_excerpt":"Most Video Large Language Models (Video-LLMs) adopt preference alignment techniques, e.g., DPO~\\citep{rafailov2024dpo}, to optimize the reward margin between a winning response ($y_w$) and a losing response ($y_l$). However, the likelihood displacement observed in DPO indicates that both $\\log \\pi_\\theta (y_w\\mid x)$ and $\\log \\pi_\\theta (y_l\\mid x) $ often decrease during training, inadvertently boosting the probabilities of non-target responses. In this paper, we systematically revisit this phenomenon from LLMs to Video-LLMs, showing that it intensifies when dealing with the redundant comple"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.05260","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/2506.05260/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":"2506.05260","created_at":"2026-07-05T11:16:44.921590+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.05260v1","created_at":"2026-07-05T11:16:44.921590+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.05260","created_at":"2026-07-05T11:16:44.921590+00:00"},{"alias_kind":"pith_short_12","alias_value":"P5I3RYXPJ7EZ","created_at":"2026-07-05T11:16:44.921590+00:00"},{"alias_kind":"pith_short_16","alias_value":"P5I3RYXPJ7EZMWSY","created_at":"2026-07-05T11:16:44.921590+00:00"},{"alias_kind":"pith_short_8","alias_value":"P5I3RYXP","created_at":"2026-07-05T11:16:44.921590+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/P5I3RYXPJ7EZMWSYUQRJPJK3PD","json":"https://pith.science/pith/P5I3RYXPJ7EZMWSYUQRJPJK3PD.json","graph_json":"https://pith.science/api/pith-number/P5I3RYXPJ7EZMWSYUQRJPJK3PD/graph.json","events_json":"https://pith.science/api/pith-number/P5I3RYXPJ7EZMWSYUQRJPJK3PD/events.json","paper":"https://pith.science/paper/P5I3RYXP"},"agent_actions":{"view_html":"https://pith.science/pith/P5I3RYXPJ7EZMWSYUQRJPJK3PD","download_json":"https://pith.science/pith/P5I3RYXPJ7EZMWSYUQRJPJK3PD.json","view_paper":"https://pith.science/paper/P5I3RYXP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.05260&json=true","fetch_graph":"https://pith.science/api/pith-number/P5I3RYXPJ7EZMWSYUQRJPJK3PD/graph.json","fetch_events":"https://pith.science/api/pith-number/P5I3RYXPJ7EZMWSYUQRJPJK3PD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/P5I3RYXPJ7EZMWSYUQRJPJK3PD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/P5I3RYXPJ7EZMWSYUQRJPJK3PD/action/storage_attestation","attest_author":"https://pith.science/pith/P5I3RYXPJ7EZMWSYUQRJPJK3PD/action/author_attestation","sign_citation":"https://pith.science/pith/P5I3RYXPJ7EZMWSYUQRJPJK3PD/action/citation_signature","submit_replication":"https://pith.science/pith/P5I3RYXPJ7EZMWSYUQRJPJK3PD/action/replication_record"}},"created_at":"2026-07-05T11:16:44.921590+00:00","updated_at":"2026-07-05T11:16:44.921590+00:00"}