{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:AQSYIM3JF3ODVDTNZYIDIEJMQJ","short_pith_number":"pith:AQSYIM3J","schema_version":"1.0","canonical_sha256":"04258433692edc3a8e6dce1034112c82626dbd6088028fa118db774711a5becb","source":{"kind":"arxiv","id":"2502.12635","version":3},"attestation_state":"computed","paper":{"title":"Corrupted but Not Broken: Understanding and Mitigating the Negative Impacts of Corrupted Data in Visual Instruction Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bo Han, Hansi Yang, James T. Kwok, Kai Chen, Lanqing Hong, Qun Liu, Yihan Zeng, Yunhao Gou, Yu Zhang, Zhenguo Li, Zhili Liu","submitted_at":"2025-02-18T08:28:29Z","abstract_excerpt":"Visual Instruction Tuning (VIT) aims to enhance Multimodal Large Language Models (MLLMs), yet its effectiveness is often compromised by corrupted datasets with issues such as hallucinated content, incorrect responses, and poor OCR quality. Previous approaches to address these challenges have focused on refining datasets through high-quality data collection or rule-based filtering that can be costly or limited in scope. In this paper, we conduct a systematic investigation into the impact of corrupted data on MLLMs and discover that, although corrupted data degrade model performance, such advers"},"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":"2502.12635","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-02-18T08:28:29Z","cross_cats_sorted":[],"title_canon_sha256":"51f5c7b738d705749fd33e64dccc2ec047fbeac437149bfb0575963cfb99314b","abstract_canon_sha256":"ae9d3b50f0ee70372bdb1636d0c13b6b43707b4ad3eac18a3c5da00796574df0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:10:05.051768Z","signature_b64":"MzDTtkutQriVqf56XfrGCjWA4WOvFyVW+L0dQPppvARg/BNCEdsmPibEmEU6gGPhrOQhz2jgV+lHaVbIpR9NBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"04258433692edc3a8e6dce1034112c82626dbd6088028fa118db774711a5becb","last_reissued_at":"2026-07-05T11:10:05.051241Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:10:05.051241Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Corrupted but Not Broken: Understanding and Mitigating the Negative Impacts of Corrupted Data in Visual Instruction Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bo Han, Hansi Yang, James T. Kwok, Kai Chen, Lanqing Hong, Qun Liu, Yihan Zeng, Yunhao Gou, Yu Zhang, Zhenguo Li, Zhili Liu","submitted_at":"2025-02-18T08:28:29Z","abstract_excerpt":"Visual Instruction Tuning (VIT) aims to enhance Multimodal Large Language Models (MLLMs), yet its effectiveness is often compromised by corrupted datasets with issues such as hallucinated content, incorrect responses, and poor OCR quality. Previous approaches to address these challenges have focused on refining datasets through high-quality data collection or rule-based filtering that can be costly or limited in scope. In this paper, we conduct a systematic investigation into the impact of corrupted data on MLLMs and discover that, although corrupted data degrade model performance, such advers"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.12635","kind":"arxiv","version":3},"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/2502.12635/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":"2502.12635","created_at":"2026-07-05T11:10:05.051304+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.12635v3","created_at":"2026-07-05T11:10:05.051304+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.12635","created_at":"2026-07-05T11:10:05.051304+00:00"},{"alias_kind":"pith_short_12","alias_value":"AQSYIM3JF3OD","created_at":"2026-07-05T11:10:05.051304+00:00"},{"alias_kind":"pith_short_16","alias_value":"AQSYIM3JF3ODVDTN","created_at":"2026-07-05T11:10:05.051304+00:00"},{"alias_kind":"pith_short_8","alias_value":"AQSYIM3J","created_at":"2026-07-05T11:10:05.051304+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.01735","citing_title":"ECCV 2024 W-CODA: 1st Workshop on Multimodal Perception and Comprehension of Corner Cases in Autonomous Driving","ref_index":15,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AQSYIM3JF3ODVDTNZYIDIEJMQJ","json":"https://pith.science/pith/AQSYIM3JF3ODVDTNZYIDIEJMQJ.json","graph_json":"https://pith.science/api/pith-number/AQSYIM3JF3ODVDTNZYIDIEJMQJ/graph.json","events_json":"https://pith.science/api/pith-number/AQSYIM3JF3ODVDTNZYIDIEJMQJ/events.json","paper":"https://pith.science/paper/AQSYIM3J"},"agent_actions":{"view_html":"https://pith.science/pith/AQSYIM3JF3ODVDTNZYIDIEJMQJ","download_json":"https://pith.science/pith/AQSYIM3JF3ODVDTNZYIDIEJMQJ.json","view_paper":"https://pith.science/paper/AQSYIM3J","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.12635&json=true","fetch_graph":"https://pith.science/api/pith-number/AQSYIM3JF3ODVDTNZYIDIEJMQJ/graph.json","fetch_events":"https://pith.science/api/pith-number/AQSYIM3JF3ODVDTNZYIDIEJMQJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AQSYIM3JF3ODVDTNZYIDIEJMQJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AQSYIM3JF3ODVDTNZYIDIEJMQJ/action/storage_attestation","attest_author":"https://pith.science/pith/AQSYIM3JF3ODVDTNZYIDIEJMQJ/action/author_attestation","sign_citation":"https://pith.science/pith/AQSYIM3JF3ODVDTNZYIDIEJMQJ/action/citation_signature","submit_replication":"https://pith.science/pith/AQSYIM3JF3ODVDTNZYIDIEJMQJ/action/replication_record"}},"created_at":"2026-07-05T11:10:05.051304+00:00","updated_at":"2026-07-05T11:10:05.051304+00:00"}