{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2023:GDV46QHWLB6AGJVNDS7TLFCWY2","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"005fb17ca22012a217f56f74db2d7f620a1da22e2332a44702e9a4daa218764b","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2023-11-28T14:54:37Z","title_canon_sha256":"cfd50767e98148c085f71f832e663a8ccb28a0b88fa4daa98d3430463eb4a27c"},"schema_version":"1.0","source":{"id":"2311.16839","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2311.16839","created_at":"2026-05-17T23:38:14Z"},{"alias_kind":"arxiv_version","alias_value":"2311.16839v2","created_at":"2026-05-17T23:38:14Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2311.16839","created_at":"2026-05-17T23:38:14Z"},{"alias_kind":"pith_short_12","alias_value":"GDV46QHWLB6A","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_16","alias_value":"GDV46QHWLB6AGJVN","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_8","alias_value":"GDV46QHW","created_at":"2026-05-18T12:33:33Z"}],"graph_snapshots":[{"event_id":"sha256:a1087b8f58614e9e98ef174a5ad999ee3008c47875b2f91a52e788544993bef5","target":"graph","created_at":"2026-05-17T23:38:14Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"When applied to three mainstream multimodal models, HA-DPO significantly reduced hallucination issues and amplified the models' generalization capabilities. Notably, the MiniGPT-4 model, when enhanced with HA-DPO, demonstrated a substantial improvement: POPE accuracy rose from 51.13% to 86.13% (an absolute improvement of 35%), and the MME score surged from 932.00 to 1326.46 (a relative improvement of 42.32%)."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The constructed positive and negative sample pairs are high-quality, style-consistent, and free of new biases that could undermine preference learning or generalization beyond the tested benchmarks."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"HA-DPO reframes hallucination reduction in LVLMs as direct preference optimization over style-consistent positive and negative response pairs, yielding large gains such as 35-point POPE accuracy jumps on MiniGPT-4."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"HA-DPO trains multimodal models to prefer accurate image descriptions over hallucinatory ones by optimizing on paired responses."}],"snapshot_sha256":"6bcaca4640cf7cd7cb48662118152e2fd33ec01a53e36f970b9ae61704a55d89"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Multimodal large language models have made significant advancements in recent years, yet they still suffer from a common issue known as the \"hallucination problem\", in which the models generate textual descriptions that inaccurately depict or entirely fabricate content from associated images. This paper introduces a novel solution, Hallucination-Aware Direct Preference Optimization (HA-DPO), which reframes the hallucination problem as a preference selection task. The model is trained to favor the non-hallucinating response when presented with two responses of the same image (one accurate and o","authors_text":"Bin Wang, Conghui He, Jiaqi Wang, Linke Ouyang, Xiaoyi Dong, Zhiyuan Zhao","cross_cats":["cs.CL"],"headline":"HA-DPO trains multimodal models to prefer accurate image descriptions over hallucinatory ones by optimizing on paired responses.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2023-11-28T14:54:37Z","title":"Beyond Hallucinations: Enhancing LVLMs through Hallucination-Aware Direct Preference Optimization"},"references":{"count":79,"internal_anchors":1,"resolved_work":79,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Brown, Jack Clark, Sam McCandlish, Chris Olah, Benjamin Mann, and Jared Kaplan","work_id":"42c49e87-0d71-4159-82b0-335c1cdf982a","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"The exploration-exploitation dilemma: a multidisci- plinary framework","work_id":"1b85bd8c-2331-4169-adae-4e756e673fca","year":2014},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Sub- biah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakan- tan, Pranav Shyam, Girish Sastry, Amanda Askell, Sand- hini Agarwal, Ariel Herbert-V oss, ","work_id":"36ef20ad-b8cb-4b11-9b0f-2e2d164b0174","year":2020},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Christiano, Jan Leike, Tom B","work_id":"a2d819f8-2c2f-42a1-bd50-b96b93d59729","year":2017},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, and Steven C. H. Hoi. Instructblip: Towards general- purpose vision-language models wit","work_id":"ca0fe730-e9d7-436a-a761-ceba8bc2e1d3","year":2023}],"snapshot_sha256":"28bd0671d7c0b4edd136e93e111c48f18e82204dcaadae758ba18f98c8437de2"},"source":{"id":"2311.16839","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-17T10:48:14.256416Z","id":"0a33879e-fd3f-4015-b277-eaf4122f1c90","model_set":{"reader":"grok-4.3"},"one_line_summary":"HA-DPO reframes hallucination reduction in LVLMs as direct preference optimization over style-consistent positive and negative response pairs, yielding large gains such as 35-point POPE accuracy jumps on MiniGPT-4.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"HA-DPO trains multimodal models to prefer accurate image descriptions over hallucinatory ones by optimizing on paired responses.","strongest_claim":"When applied to three mainstream multimodal models, HA-DPO significantly reduced hallucination issues and amplified the models' generalization capabilities. Notably, the MiniGPT-4 model, when enhanced with HA-DPO, demonstrated a substantial improvement: POPE accuracy rose from 51.13% to 86.13% (an absolute improvement of 35%), and the MME score surged from 932.00 to 1326.46 (a relative improvement of 42.32%).","weakest_assumption":"The constructed positive and negative sample pairs are high-quality, style-consistent, and free of new biases that could undermine preference learning or generalization beyond the tested benchmarks."}},"verdict_id":"0a33879e-fd3f-4015-b277-eaf4122f1c90"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:d6f3a9e9616ddc9c64b0eeba69aa686a7425a7088a1289c27934461a810dacea","target":"record","created_at":"2026-05-17T23:38:14Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"005fb17ca22012a217f56f74db2d7f620a1da22e2332a44702e9a4daa218764b","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2023-11-28T14:54:37Z","title_canon_sha256":"cfd50767e98148c085f71f832e663a8ccb28a0b88fa4daa98d3430463eb4a27c"},"schema_version":"1.0","source":{"id":"2311.16839","kind":"arxiv","version":2}},"canonical_sha256":"30ebcf40f6587c0326ad1cbf359456c6a539c274676e43cc41abba0df6f8f6da","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"30ebcf40f6587c0326ad1cbf359456c6a539c274676e43cc41abba0df6f8f6da","first_computed_at":"2026-05-17T23:38:14.312573Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:14.312573Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"qNsy8/9mRDSA3l9DFns72Xxn5Rzd5LIr0zuT9eMgKndUU9ZGvW1ht5sWh9lsh5ZGSZBwAvChHdswNQ38OxbeAA==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:14.313144Z","signed_message":"canonical_sha256_bytes"},"source_id":"2311.16839","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:d6f3a9e9616ddc9c64b0eeba69aa686a7425a7088a1289c27934461a810dacea","sha256:a1087b8f58614e9e98ef174a5ad999ee3008c47875b2f91a52e788544993bef5"],"state_sha256":"197ffcee8d5ad5a1379848d1fa8b85834c38e56073999be37a339bacb0e82ae2"}