{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:YS7P6QJYSIKZXE6QTNT6E2WBCQ","short_pith_number":"pith:YS7P6QJY","schema_version":"1.0","canonical_sha256":"c4beff413892159b93d09b67e26ac11426a3320caec4ac7cda7a366ac8a83126","source":{"kind":"arxiv","id":"2503.03122","version":4},"attestation_state":"computed","paper":{"title":"The Devil Is in the Details: Tackling Unimodal Spurious Correlations for Generalizable Multimodal Reward Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Debing Zhang, Jie Lou, Le Sun, Xianpei Han, Xueru Wen, Yaojie Lu, Yuqiu Ji, Zichao Li","submitted_at":"2025-03-05T02:37:41Z","abstract_excerpt":"Multimodal Reward Models (MM-RMs) are crucial for aligning Large Language Models (LLMs) with human preferences, particularly as LLMs increasingly interact with multimodal data. However, we find that MM-RMs trained on existing datasets often struggle to generalize to out-of-distribution data due to their reliance on unimodal spurious correlations, primarily text-only shortcuts within the training distribution, which prevents them from leveraging true multimodal reward functions. To address this, we introduce a Shortcut-aware MM-RM learning algorithm that mitigates this issue by dynamically rewe"},"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":"2503.03122","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-03-05T02:37:41Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"2de809dbaa6e64b18eddd4b62ba4b11b953e992554092af5f7e38a54da13f92f","abstract_canon_sha256":"c6e9b12a4c9d8f42489960b978d5f3cddf68c789b7c48faf25a26dcc979d773d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:06:29.182412Z","signature_b64":"IoSIKPcEU9T76BnygcygrwRjDOWBDASOgNiVBAOM1q6T+2PICQbppg+WOVkWnUR4b/SjUzY5ZknR7A2XXZrQDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c4beff413892159b93d09b67e26ac11426a3320caec4ac7cda7a366ac8a83126","last_reissued_at":"2026-07-05T11:06:29.181903Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:06:29.181903Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Devil Is in the Details: Tackling Unimodal Spurious Correlations for Generalizable Multimodal Reward Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Debing Zhang, Jie Lou, Le Sun, Xianpei Han, Xueru Wen, Yaojie Lu, Yuqiu Ji, Zichao Li","submitted_at":"2025-03-05T02:37:41Z","abstract_excerpt":"Multimodal Reward Models (MM-RMs) are crucial for aligning Large Language Models (LLMs) with human preferences, particularly as LLMs increasingly interact with multimodal data. However, we find that MM-RMs trained on existing datasets often struggle to generalize to out-of-distribution data due to their reliance on unimodal spurious correlations, primarily text-only shortcuts within the training distribution, which prevents them from leveraging true multimodal reward functions. To address this, we introduce a Shortcut-aware MM-RM learning algorithm that mitigates this issue by dynamically rewe"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.03122","kind":"arxiv","version":4},"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/2503.03122/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":"2503.03122","created_at":"2026-07-05T11:06:29.181964+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.03122v4","created_at":"2026-07-05T11:06:29.181964+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.03122","created_at":"2026-07-05T11:06:29.181964+00:00"},{"alias_kind":"pith_short_12","alias_value":"YS7P6QJYSIKZ","created_at":"2026-07-05T11:06:29.181964+00:00"},{"alias_kind":"pith_short_16","alias_value":"YS7P6QJYSIKZXE6Q","created_at":"2026-07-05T11:06:29.181964+00:00"},{"alias_kind":"pith_short_8","alias_value":"YS7P6QJY","created_at":"2026-07-05T11:06:29.181964+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.25343","citing_title":"Toward Native Multimodal Modeling: A Roadmap","ref_index":189,"is_internal_anchor":false},{"citing_arxiv_id":"2505.07062","citing_title":"Seed1.5-VL Technical Report","ref_index":76,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YS7P6QJYSIKZXE6QTNT6E2WBCQ","json":"https://pith.science/pith/YS7P6QJYSIKZXE6QTNT6E2WBCQ.json","graph_json":"https://pith.science/api/pith-number/YS7P6QJYSIKZXE6QTNT6E2WBCQ/graph.json","events_json":"https://pith.science/api/pith-number/YS7P6QJYSIKZXE6QTNT6E2WBCQ/events.json","paper":"https://pith.science/paper/YS7P6QJY"},"agent_actions":{"view_html":"https://pith.science/pith/YS7P6QJYSIKZXE6QTNT6E2WBCQ","download_json":"https://pith.science/pith/YS7P6QJYSIKZXE6QTNT6E2WBCQ.json","view_paper":"https://pith.science/paper/YS7P6QJY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.03122&json=true","fetch_graph":"https://pith.science/api/pith-number/YS7P6QJYSIKZXE6QTNT6E2WBCQ/graph.json","fetch_events":"https://pith.science/api/pith-number/YS7P6QJYSIKZXE6QTNT6E2WBCQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YS7P6QJYSIKZXE6QTNT6E2WBCQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YS7P6QJYSIKZXE6QTNT6E2WBCQ/action/storage_attestation","attest_author":"https://pith.science/pith/YS7P6QJYSIKZXE6QTNT6E2WBCQ/action/author_attestation","sign_citation":"https://pith.science/pith/YS7P6QJYSIKZXE6QTNT6E2WBCQ/action/citation_signature","submit_replication":"https://pith.science/pith/YS7P6QJYSIKZXE6QTNT6E2WBCQ/action/replication_record"}},"created_at":"2026-07-05T11:06:29.181964+00:00","updated_at":"2026-07-05T11:06:29.181964+00:00"}