{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:IGRTTSZZPTLWU75YMQBIWSXVGP","short_pith_number":"pith:IGRTTSZZ","schema_version":"1.0","canonical_sha256":"41a339cb397cd76a7fb864028b4af533d69a3ba7dbe7fb1a404a4bb77ac41be1","source":{"kind":"arxiv","id":"2607.28590","version":1},"attestation_state":"computed","paper":{"title":"VAD: Attributing Visual Evidence for Target Reconstruction in Multimodal On-Policy Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CV","authors_text":"Jianghao Lin, Kangning Zhang, Qingyao Li, Shuai Shao, Weinan Zhang, Weiwen Liu, Wenxiang Jiao, Yixing Li, Yong Yu, Yuan Lu, Zhengxi Lu, Zhiyuan Yao","submitted_at":"2026-07-30T17:43:46Z","abstract_excerpt":"Multimodal on-policy distillation (OPD) transfers fine-grained visual knowledge by supervising student-generated trajectories with a privileged-view teacher. Yet its next-token corrections are source-mixed, combining visual signals with linguistic priors and teacher-specific effects. The key challenge is to estimate which corrections are supported by visual evidence, not merely where or how strongly to distill. We introduce Visual Attribution Distillation (VAD), a counterfactual target-reconstruction algorithm that estimates the visually attributable part of a teacher correction. At each stude"},"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":"2607.28590","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-07-30T17:43:46Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"410c0687a44fec0905df2a11367cfbf4fd14fc9cc5dc6d936107c1309acc9dd9","abstract_canon_sha256":"6120ad4268f8e0e6d9e6efbcac1372a581ff0d32f5ca1c4068b0ebddb7539f40"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"41a339cb397cd76a7fb864028b4af533d69a3ba7dbe7fb1a404a4bb77ac41be1","last_reissued_at":"2026-07-31T01:38:47.861932Z","signature_status":"unsigned_v0","first_computed_at":"2026-07-31T01:38:47.861932Z"},"graph_snapshot":{"paper":{"title":"VAD: Attributing Visual Evidence for Target Reconstruction in Multimodal On-Policy Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CV","authors_text":"Jianghao Lin, Kangning Zhang, Qingyao Li, Shuai Shao, Weinan Zhang, Weiwen Liu, Wenxiang Jiao, Yixing Li, Yong Yu, Yuan Lu, Zhengxi Lu, Zhiyuan Yao","submitted_at":"2026-07-30T17:43:46Z","abstract_excerpt":"Multimodal on-policy distillation (OPD) transfers fine-grained visual knowledge by supervising student-generated trajectories with a privileged-view teacher. Yet its next-token corrections are source-mixed, combining visual signals with linguistic priors and teacher-specific effects. The key challenge is to estimate which corrections are supported by visual evidence, not merely where or how strongly to distill. We introduce Visual Attribution Distillation (VAD), a counterfactual target-reconstruction algorithm that estimates the visually attributable part of a teacher correction. At each stude"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.28590","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/2607.28590/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":"2607.28590","created_at":"2026-07-31T01:38:47.865182+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.28590v1","created_at":"2026-07-31T01:38:47.865182+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.28590","created_at":"2026-07-31T01:38:47.865182+00:00"},{"alias_kind":"pith_short_12","alias_value":"IGRTTSZZPTLW","created_at":"2026-07-31T01:38:47.865182+00:00"},{"alias_kind":"pith_short_16","alias_value":"IGRTTSZZPTLWU75Y","created_at":"2026-07-31T01:38:47.865182+00:00"},{"alias_kind":"pith_short_8","alias_value":"IGRTTSZZ","created_at":"2026-07-31T01:38:47.865182+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2608.11980","citing_title":"Learning from Unreachable Rewards: Hint-Conditioned Reinforcement Learning for Generative Recommendation","ref_index":2026,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IGRTTSZZPTLWU75YMQBIWSXVGP","json":"https://pith.science/pith/IGRTTSZZPTLWU75YMQBIWSXVGP.json","graph_json":"https://pith.science/api/pith-number/IGRTTSZZPTLWU75YMQBIWSXVGP/graph.json","events_json":"https://pith.science/api/pith-number/IGRTTSZZPTLWU75YMQBIWSXVGP/events.json","paper":"https://pith.science/paper/IGRTTSZZ"},"agent_actions":{"view_html":"https://pith.science/pith/IGRTTSZZPTLWU75YMQBIWSXVGP","download_json":"https://pith.science/pith/IGRTTSZZPTLWU75YMQBIWSXVGP.json","view_paper":"https://pith.science/paper/IGRTTSZZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.28590&json=true","fetch_graph":"https://pith.science/api/pith-number/IGRTTSZZPTLWU75YMQBIWSXVGP/graph.json","fetch_events":"https://pith.science/api/pith-number/IGRTTSZZPTLWU75YMQBIWSXVGP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IGRTTSZZPTLWU75YMQBIWSXVGP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IGRTTSZZPTLWU75YMQBIWSXVGP/action/storage_attestation","attest_author":"https://pith.science/pith/IGRTTSZZPTLWU75YMQBIWSXVGP/action/author_attestation","sign_citation":"https://pith.science/pith/IGRTTSZZPTLWU75YMQBIWSXVGP/action/citation_signature","submit_replication":"https://pith.science/pith/IGRTTSZZPTLWU75YMQBIWSXVGP/action/replication_record"}},"created_at":"2026-07-31T01:38:47.865182+00:00","updated_at":"2026-07-31T01:38:47.865182+00:00"}