{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:FPFRXLS7LJQYPWA6CKJ6FK6FNT","short_pith_number":"pith:FPFRXLS7","schema_version":"1.0","canonical_sha256":"2bcb1bae5f5a6187d81e1293e2abc56cc4d34bc7df168af78d83a07a1206e8b7","source":{"kind":"arxiv","id":"2502.00577","version":2},"attestation_state":"computed","paper":{"title":"Understanding Multimodal LLMs Under Distribution Shifts: An Information-Theoretic Approach","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Changdae Oh, Shawn Im, Xuefeng Du, Yixuan Li, Zhen Fang","submitted_at":"2025-02-01T22:06:56Z","abstract_excerpt":"Multimodal large language models (MLLMs) have shown promising capabilities but struggle under distribution shifts, where evaluation data differ from instruction tuning distributions. Although previous works have provided empirical evaluations, we argue that establishing a formal framework that can characterize and quantify the risk of MLLMs is necessary to ensure the safe and reliable application of MLLMs in the real world. By taking an information-theoretic perspective, we propose the first theoretical framework that enables the quantification of the maximum risk of MLLMs under distribution s"},"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":"2502.00577","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-02-01T22:06:56Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"f6ddf924ad9bca4bd4aeed138f69a81e0fc259f828a22e79a5c79850280dcf99","abstract_canon_sha256":"f4729edf26b88c51bd27d9ed23b82805a4a2d310cc9fb1d723fa1cd88a3235bb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:09:07.427567Z","signature_b64":"ZPxrGc1KwG0rKnt4YWNu52KalgZz+NDd5DDsBsuEMf2AtNokWLXh7nhwTD8+2MFRjoh0byd8LsVkpiaShohCDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2bcb1bae5f5a6187d81e1293e2abc56cc4d34bc7df168af78d83a07a1206e8b7","last_reissued_at":"2026-07-05T11:09:07.427079Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:09:07.427079Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding Multimodal LLMs Under Distribution Shifts: An Information-Theoretic Approach","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Changdae Oh, Shawn Im, Xuefeng Du, Yixuan Li, Zhen Fang","submitted_at":"2025-02-01T22:06:56Z","abstract_excerpt":"Multimodal large language models (MLLMs) have shown promising capabilities but struggle under distribution shifts, where evaluation data differ from instruction tuning distributions. Although previous works have provided empirical evaluations, we argue that establishing a formal framework that can characterize and quantify the risk of MLLMs is necessary to ensure the safe and reliable application of MLLMs in the real world. By taking an information-theoretic perspective, we propose the first theoretical framework that enables the quantification of the maximum risk of MLLMs under distribution s"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.00577","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/2502.00577/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":"2502.00577","created_at":"2026-07-05T11:09:07.427137+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.00577v2","created_at":"2026-07-05T11:09:07.427137+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.00577","created_at":"2026-07-05T11:09:07.427137+00:00"},{"alias_kind":"pith_short_12","alias_value":"FPFRXLS7LJQY","created_at":"2026-07-05T11:09:07.427137+00:00"},{"alias_kind":"pith_short_16","alias_value":"FPFRXLS7LJQYPWA6","created_at":"2026-07-05T11:09:07.427137+00:00"},{"alias_kind":"pith_short_8","alias_value":"FPFRXLS7","created_at":"2026-07-05T11:09:07.427137+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.20720","citing_title":"COMPASS: COntinual Multilingual PEFT with Adaptive Semantic Sampling","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19488","citing_title":"CoDA: Towards Effective Cross-domain Knowledge Transfer via CoT-guided Domain Adaptation","ref_index":57,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FPFRXLS7LJQYPWA6CKJ6FK6FNT","json":"https://pith.science/pith/FPFRXLS7LJQYPWA6CKJ6FK6FNT.json","graph_json":"https://pith.science/api/pith-number/FPFRXLS7LJQYPWA6CKJ6FK6FNT/graph.json","events_json":"https://pith.science/api/pith-number/FPFRXLS7LJQYPWA6CKJ6FK6FNT/events.json","paper":"https://pith.science/paper/FPFRXLS7"},"agent_actions":{"view_html":"https://pith.science/pith/FPFRXLS7LJQYPWA6CKJ6FK6FNT","download_json":"https://pith.science/pith/FPFRXLS7LJQYPWA6CKJ6FK6FNT.json","view_paper":"https://pith.science/paper/FPFRXLS7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.00577&json=true","fetch_graph":"https://pith.science/api/pith-number/FPFRXLS7LJQYPWA6CKJ6FK6FNT/graph.json","fetch_events":"https://pith.science/api/pith-number/FPFRXLS7LJQYPWA6CKJ6FK6FNT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FPFRXLS7LJQYPWA6CKJ6FK6FNT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FPFRXLS7LJQYPWA6CKJ6FK6FNT/action/storage_attestation","attest_author":"https://pith.science/pith/FPFRXLS7LJQYPWA6CKJ6FK6FNT/action/author_attestation","sign_citation":"https://pith.science/pith/FPFRXLS7LJQYPWA6CKJ6FK6FNT/action/citation_signature","submit_replication":"https://pith.science/pith/FPFRXLS7LJQYPWA6CKJ6FK6FNT/action/replication_record"}},"created_at":"2026-07-05T11:09:07.427137+00:00","updated_at":"2026-07-05T11:09:07.427137+00:00"}