{"id":"56418bb94a53143cd01ef6cbecb892a9abb0eeb7d58300ea29df2777b1105055","title":"arXiv cs.AI: TIB: Sample-wise Tempered Information Bottleneck for Multimodal Attribution beyond Alignment Assumption","argument":"Authors Qiming Huang, Peixi Liu, and Jianbo Jiao argue that multimodal attribution can go wrong when image-text pairs are treated as equally reliable evidence, even though some pairs are only partly aligned in meaning. Multimodal attribution seeks to interpret a model without ground-truth explanatory supervision; the paper says existing methods often use paired modalities as proxy supervision, assuming semantic alignment. The authors’ theoretical analysis says that, when this assumption fails, Information bottleneck (IB)-based methods can produce “posterior over-contraction and forced explanations”: in other words, the method may constrain its explanations too much and force an explanation from unreliable pairings. They propose TIB, which adjusts attribution strength sample by sample according to cross-modal reliability, without assuming semantic alignment. The authors report experiments on controlled benchmarks and large-scale datasets that they say show TIB’s effectiveness and validate the identified failure mode. The abstract does not give benchmark details or results, so the extent of that evidence cannot be assessed from this page alone.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract does not specify the controlled benchmarks or large-scale datasets, report results, or describe how cross-modal reliability is measured. The claims about theoretical analysis and experimental effectiveness are the authors’ claims as presented in the abstract; the supplied page does not include the paper’s supporting details.","speaker":{"name":"Huang, Qiming, Liu, Peixi, Jiao, Jianbo","roleAtSource":"Article author"},"source":{"title":"TIB: Sample-wise Tempered Information Bottleneck for Multimodal Attribution beyond Alignment Assumption","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.12997","publishedAt":"2026-11-08","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors say existing multimodal attribution methods commonly use paired modalities as proxy supervision and assume image-text pairs are semantically aligned.","excerpt":"Existing multimodal attribution methods typically rely on paired modalities as proxy supervision, implicitly assuming that image-text pairs are semantically aligned.","url":"https://arxiv.org/abs/2611.12997","relationship":"Source-stated"},{"text":"The authors’ theoretical analysis says misalignment can lead Information bottleneck-based attribution objectives to over-constrain their posterior and force explanations.","excerpt":"they induce posterior over-contraction and forced explanations.","url":"https://arxiv.org/abs/2611.12997","relationship":"Source-stated"},{"text":"The authors propose TIB, which adapts attribution strength to cross-modal reliability without assuming semantic alignment.","excerpt":"adaptively modulates attribution strength according to cross-modal reliability, without assuming semantic alignment.","url":"https://arxiv.org/abs/2611.12997","relationship":"Source-stated"},{"text":"The authors report experiments on controlled benchmarks and large-scale datasets as evidence for TIB’s effectiveness and the failure mode they identify.","excerpt":"Extensive experiments on controlled benchmarks and large-scale datasets show the effectiveness of the proposed approach, while also validating the identified failure mode.","url":"https://arxiv.org/abs/2611.12997","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.12997","checkedAt":"2026-11-11T20:58:14.483Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How do the full paper’s controlled benchmarks and large-scale datasets measure TIB’s effectiveness when cross-modal reliability varies?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: TIB: Sample-wise Tempered Information Bottleneck for Multimodal Attribution beyond Alignment Assumption","summary":"Authors Qiming Huang, Peixi Liu, and Jianbo Jiao argue that multimodal attribution can go wrong when image-text pairs are treated as equally reliable evidence, even though some pairs are only partly aligned in meaning. Multimodal attribution seeks to interpret a model without ground-truth explanatory supervision; the paper says existing methods often use paired modalities as proxy supervision, assuming semantic alignment. The authors’ theoretical analysis says that, when this assumption fails, Information bottleneck (IB)-based methods can produce “posterior over-contraction and forced explanations”: in other words, the method may constrain its explanations too much and force an explanation from unreliable pairings. They propose TIB, which adjusts attribution strength sample by sample according to cross-modal reliability, without assuming semantic alignment. The authors report experiments on controlled benchmarks and large-scale datasets that they say show TIB’s effectiveness and validate the identified failure mode. The abstract does not give benchmark details or results, so the extent of that evidence cannot be assessed from this page alone.","question":"How do the full paper’s controlled benchmarks and large-scale datasets measure TIB’s effectiveness when cross-modal reliability varies?","claims":[{"text":"The authors say existing multimodal attribution methods commonly use paired modalities as proxy supervision and assume image-text pairs are semantically aligned.","excerpt":"Existing multimodal attribution methods typically rely on paired modalities as proxy supervision, implicitly assuming that image-text pairs are semantically aligned.","support":"source-stated"},{"text":"The authors’ theoretical analysis says misalignment can lead Information bottleneck-based attribution objectives to over-constrain their posterior and force explanations.","excerpt":"they induce posterior over-contraction and forced explanations.","support":"source-stated"},{"text":"The authors propose TIB, which adapts attribution strength to cross-modal reliability without assuming semantic alignment.","excerpt":"adaptively modulates attribution strength according to cross-modal reliability, without assuming semantic alignment.","support":"source-stated"},{"text":"The authors report experiments on controlled benchmarks and large-scale datasets as evidence for TIB’s effectiveness and the failure mode they identify.","excerpt":"Extensive experiments on controlled benchmarks and large-scale datasets show the effectiveness of the proposed approach, while also validating the identified failure mode.","support":"source-stated"}],"limitations":["The abstract does not specify the controlled benchmarks or large-scale datasets, report results, or describe how cross-modal reliability is measured.","The claims about theoretical analysis and experimental effectiveness are the authors’ claims as presented in the abstract; the supplied page does not include the paper’s supporting details."],"mediaId":"8819b80ec5833832099316f8b941e46881743391b16dc31edf3ffad95c0e8ee4","correctionReason":null,"id":"56418bb94a53143cd01ef6cbecb892a9abb0eeb7d58300ea29df2777b1105055","publicationId":"c111a40f-bbec-4157-8648-1d46e8ba36ca","version":1,"sourceUrl":"https://arxiv.org/abs/2611.12997","sourceTitle":"TIB: Sample-wise Tempered Information Bottleneck for Multimodal Attribution beyond Alignment Assumption","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-08","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-11T05:01:26.428Z","createdAt":"2026-11-11T18:50:19.620Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"8819b80ec5833832099316f8b941e46881743391b16dc31edf3ffad95c0e8ee4","kind":"article","url":"https://arxiv.org/abs/2611.12997","sourceUrl":"https://arxiv.org/abs/2611.12997","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-11T18:49:10.277Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.12997","checkedAt":"2026-11-11T20:58:14.483Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"8819b80ec5833832099316f8b941e46881743391b16dc31edf3ffad95c0e8ee4","confidence":0.75,"provider":"typesafe","model":"jev-1.13.0","inputHash":"43fd757bfdb982e249d9aa860e701277e9c7b2346b7d6fb9f6d0fbc48f007194","decidedAt":"2026-11-11T18:49:12.427Z"},"provenance":{"textHash":"867b41dd52bbf7e885db58e3c584098944a304ac93caf3b6fb3f2060070b6e70","observationId":"055c31d5effb5f76f8b3755d813392e6b33a5481331c1f6ba6c80cdd9abe6c0e","retrievedAt":"2026-11-11T18:49:09.950Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"538f4bf5-a4ac-43a3-836f-39c8d8b940a6"},"reviewer":{"model":"gpt-6-luna","runId":"91c0ba58-a8d9-45f8-8f28-d3615c3ee1ca"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-11T18:50:19.620Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Huang, Qiming","url":null,"kind":"unknown"},{"name":"Liu, Peixi","url":null,"kind":"unknown"},{"name":"Jiao, Jianbo","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Huang, Qiming","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Huang, Qiming","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12997","locator":"head/meta[name=author|citation_author]"},"assertionId":"2d1203a376298200a7aa8bab65a44838e8a88607498d17adc3238bc81dbb1538:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:58:14.483Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"2d1203a376298200a7aa8bab65a44838e8a88607498d17adc3238bc81dbb1538","sourceUrl":"https://arxiv.org/abs/2611.12997","contentSha256":"05b30dba7686f280380887b03032dd695d566a77e570b52017ebaa206c9c4a6d","textHash":"867b41dd52bbf7e885db58e3c584098944a304ac93caf3b6fb3f2060070b6e70","observedAt":"2026-11-11T20:58:14.483Z"}},{"actor":{"name":"Liu, Peixi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Liu, Peixi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12997","locator":"head/meta[name=author|citation_author]"},"assertionId":"2d1203a376298200a7aa8bab65a44838e8a88607498d17adc3238bc81dbb1538:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:58:14.483Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"2d1203a376298200a7aa8bab65a44838e8a88607498d17adc3238bc81dbb1538","sourceUrl":"https://arxiv.org/abs/2611.12997","contentSha256":"05b30dba7686f280380887b03032dd695d566a77e570b52017ebaa206c9c4a6d","textHash":"867b41dd52bbf7e885db58e3c584098944a304ac93caf3b6fb3f2060070b6e70","observedAt":"2026-11-11T20:58:14.483Z"}},{"actor":{"name":"Jiao, Jianbo","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Jiao, Jianbo","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12997","locator":"head/meta[name=author|citation_author]"},"assertionId":"2d1203a376298200a7aa8bab65a44838e8a88607498d17adc3238bc81dbb1538:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:58:14.483Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"2d1203a376298200a7aa8bab65a44838e8a88607498d17adc3238bc81dbb1538","sourceUrl":"https://arxiv.org/abs/2611.12997","contentSha256":"05b30dba7686f280380887b03032dd695d566a77e570b52017ebaa206c9c4a6d","textHash":"867b41dd52bbf7e885db58e3c584098944a304ac93caf3b6fb3f2060070b6e70","observedAt":"2026-11-11T20:58:14.483Z"}}],"participationCoverage":{"total":3,"returned":3,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.12997","snapshotId":"2d1203a376298200a7aa8bab65a44838e8a88607498d17adc3238bc81dbb1538","contentSha256":"05b30dba7686f280380887b03032dd695d566a77e570b52017ebaa206c9c4a6d","textHash":"867b41dd52bbf7e885db58e3c584098944a304ac93caf3b6fb3f2060070b6e70","observedAt":"2026-11-11T20:58:14.483Z","basis":"publisher-page-metadata","sourceLinks":43,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/56418bb94a53143cd01ef6cbecb892a9abb0eeb7d58300ea29df2777b1105055/source"}},"canonicalUrl":"https://escapemeridian.com/readings/arxiv-cs-ai-tib-sample-wise-tempered-information-bottleneck-for-56418bb9","formats":{"html":"/readings/arxiv-cs-ai-tib-sample-wise-tempered-information-bottleneck-for-56418bb9","markdown":"/readings/arxiv-cs-ai-tib-sample-wise-tempered-information-bottleneck-for-56418bb9.md","json":"/readings/arxiv-cs-ai-tib-sample-wise-tempered-information-bottleneck-for-56418bb9.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[{"id":"term-multimodal-model","name":"Multimodal model","explanation":"A model that works with more than one kind of input or output, such as text, images and audio.","referenceUrl":"https://developers.google.com/machine-learning/glossary#multimodal-model"},{"id":"term-benchmark","name":"Benchmark","explanation":"A defined set of tests used to compare systems. Results depend on the tasks and measures selected.","referenceUrl":"https://arxiv.org/abs/2211.09110"}],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"56418bb94a53143cd01ef6cbecb892a9abb0eeb7d58300ea29df2777b1105055","revision":"c111a40f-bbec-4157-8648-1d46e8ba36ca","revisionKind":"publication-id","title":"arXiv cs.AI: TIB: Sample-wise Tempered Information Bottleneck for Multimodal Attribution beyond Alignment Assumption","url":"https://escapemeridian.com/readings/arxiv-cs-ai-tib-sample-wise-tempered-information-bottleneck-for-56418bb9"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.12997","revision":"867b41dd52bbf7e885db58e3c584098944a304ac93caf3b6fb3f2060070b6e70","revisionKind":"sha256-source-text","title":"TIB: Sample-wise Tempered Information Bottleneck for Multimodal Attribution beyond Alignment Assumption","url":"https://arxiv.org/abs/2611.12997"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter."},"revisions":[{"id":"56418bb94a53143cd01ef6cbecb892a9abb0eeb7d58300ea29df2777b1105055","title":"arXiv cs.AI: TIB: Sample-wise Tempered Information Bottleneck for Multimodal Attribution beyond Alignment Assumption","argument":"Authors Qiming Huang, Peixi Liu, and Jianbo Jiao argue that multimodal attribution can go wrong when image-text pairs are treated as equally reliable evidence, even though some pairs are only partly aligned in meaning. Multimodal attribution seeks to interpret a model without ground-truth explanatory supervision; the paper says existing methods often use paired modalities as proxy supervision, assuming semantic alignment. The authors’ theoretical analysis says that, when this assumption fails, Information bottleneck (IB)-based methods can produce “posterior over-contraction and forced explanations”: in other words, the method may constrain its explanations too much and force an explanation from unreliable pairings. They propose TIB, which adjusts attribution strength sample by sample according to cross-modal reliability, without assuming semantic alignment. The authors report experiments on controlled benchmarks and large-scale datasets that they say show TIB’s effectiveness and validate the identified failure mode. The abstract does not give benchmark details or results, so the extent of that evidence cannot be assessed from this page alone.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract does not specify the controlled benchmarks or large-scale datasets, report results, or describe how cross-modal reliability is measured. The claims about theoretical analysis and experimental effectiveness are the authors’ claims as presented in the abstract; the supplied page does not include the paper’s supporting details.","speaker":{"name":"Huang, Qiming, Liu, Peixi, Jiao, Jianbo","roleAtSource":"Article author"},"source":{"title":"TIB: Sample-wise Tempered Information Bottleneck for Multimodal Attribution beyond Alignment Assumption","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.12997","publishedAt":"2026-11-08","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors say existing multimodal attribution methods commonly use paired modalities as proxy supervision and assume image-text pairs are semantically aligned.","excerpt":"Existing multimodal attribution methods typically rely on paired modalities as proxy supervision, implicitly assuming that image-text pairs are semantically aligned.","url":"https://arxiv.org/abs/2611.12997","relationship":"Source-stated"},{"text":"The authors’ theoretical analysis says misalignment can lead Information bottleneck-based attribution objectives to over-constrain their posterior and force explanations.","excerpt":"they induce posterior over-contraction and forced explanations.","url":"https://arxiv.org/abs/2611.12997","relationship":"Source-stated"},{"text":"The authors propose TIB, which adapts attribution strength to cross-modal reliability without assuming semantic alignment.","excerpt":"adaptively modulates attribution strength according to cross-modal reliability, without assuming semantic alignment.","url":"https://arxiv.org/abs/2611.12997","relationship":"Source-stated"},{"text":"The authors report experiments on controlled benchmarks and large-scale datasets as evidence for TIB’s effectiveness and the failure mode they identify.","excerpt":"Extensive experiments on controlled benchmarks and large-scale datasets show the effectiveness of the proposed approach, while also validating the identified failure mode.","url":"https://arxiv.org/abs/2611.12997","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.12997","checkedAt":"2026-11-11T20:58:14.483Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How do the full paper’s controlled benchmarks and large-scale datasets measure TIB’s effectiveness when cross-modal reliability varies?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: TIB: Sample-wise Tempered Information Bottleneck for Multimodal Attribution beyond Alignment Assumption","summary":"Authors Qiming Huang, Peixi Liu, and Jianbo Jiao argue that multimodal attribution can go wrong when image-text pairs are treated as equally reliable evidence, even though some pairs are only partly aligned in meaning. Multimodal attribution seeks to interpret a model without ground-truth explanatory supervision; the paper says existing methods often use paired modalities as proxy supervision, assuming semantic alignment. The authors’ theoretical analysis says that, when this assumption fails, Information bottleneck (IB)-based methods can produce “posterior over-contraction and forced explanations”: in other words, the method may constrain its explanations too much and force an explanation from unreliable pairings. They propose TIB, which adjusts attribution strength sample by sample according to cross-modal reliability, without assuming semantic alignment. The authors report experiments on controlled benchmarks and large-scale datasets that they say show TIB’s effectiveness and validate the identified failure mode. The abstract does not give benchmark details or results, so the extent of that evidence cannot be assessed from this page alone.","question":"How do the full paper’s controlled benchmarks and large-scale datasets measure TIB’s effectiveness when cross-modal reliability varies?","claims":[{"text":"The authors say existing multimodal attribution methods commonly use paired modalities as proxy supervision and assume image-text pairs are semantically aligned.","excerpt":"Existing multimodal attribution methods typically rely on paired modalities as proxy supervision, implicitly assuming that image-text pairs are semantically aligned.","support":"source-stated"},{"text":"The authors’ theoretical analysis says misalignment can lead Information bottleneck-based attribution objectives to over-constrain their posterior and force explanations.","excerpt":"they induce posterior over-contraction and forced explanations.","support":"source-stated"},{"text":"The authors propose TIB, which adapts attribution strength to cross-modal reliability without assuming semantic alignment.","excerpt":"adaptively modulates attribution strength according to cross-modal reliability, without assuming semantic alignment.","support":"source-stated"},{"text":"The authors report experiments on controlled benchmarks and large-scale datasets as evidence for TIB’s effectiveness and the failure mode they identify.","excerpt":"Extensive experiments on controlled benchmarks and large-scale datasets show the effectiveness of the proposed approach, while also validating the identified failure mode.","support":"source-stated"}],"limitations":["The abstract does not specify the controlled benchmarks or large-scale datasets, report results, or describe how cross-modal reliability is measured.","The claims about theoretical analysis and experimental effectiveness are the authors’ claims as presented in the abstract; the supplied page does not include the paper’s supporting details."],"mediaId":"8819b80ec5833832099316f8b941e46881743391b16dc31edf3ffad95c0e8ee4","correctionReason":null,"id":"56418bb94a53143cd01ef6cbecb892a9abb0eeb7d58300ea29df2777b1105055","publicationId":"c111a40f-bbec-4157-8648-1d46e8ba36ca","version":1,"sourceUrl":"https://arxiv.org/abs/2611.12997","sourceTitle":"TIB: Sample-wise Tempered Information Bottleneck for Multimodal Attribution beyond Alignment Assumption","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-08","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-11T05:01:26.428Z","createdAt":"2026-11-11T18:50:19.620Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"8819b80ec5833832099316f8b941e46881743391b16dc31edf3ffad95c0e8ee4","kind":"article","url":"https://arxiv.org/abs/2611.12997","sourceUrl":"https://arxiv.org/abs/2611.12997","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-11T18:49:10.277Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.12997","checkedAt":"2026-11-11T20:58:14.483Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"8819b80ec5833832099316f8b941e46881743391b16dc31edf3ffad95c0e8ee4","confidence":0.75,"provider":"typesafe","model":"jev-1.13.0","inputHash":"43fd757bfdb982e249d9aa860e701277e9c7b2346b7d6fb9f6d0fbc48f007194","decidedAt":"2026-11-11T18:49:12.427Z"},"provenance":{"textHash":"867b41dd52bbf7e885db58e3c584098944a304ac93caf3b6fb3f2060070b6e70","observationId":"055c31d5effb5f76f8b3755d813392e6b33a5481331c1f6ba6c80cdd9abe6c0e","retrievedAt":"2026-11-11T18:49:09.950Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"538f4bf5-a4ac-43a3-836f-39c8d8b940a6"},"reviewer":{"model":"gpt-6-luna","runId":"91c0ba58-a8d9-45f8-8f28-d3615c3ee1ca"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-11T18:50:19.620Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Huang, Qiming","url":null,"kind":"unknown"},{"name":"Liu, Peixi","url":null,"kind":"unknown"},{"name":"Jiao, Jianbo","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Huang, Qiming","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Huang, Qiming","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12997","locator":"head/meta[name=author|citation_author]"},"assertionId":"2d1203a376298200a7aa8bab65a44838e8a88607498d17adc3238bc81dbb1538:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:58:14.483Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"2d1203a376298200a7aa8bab65a44838e8a88607498d17adc3238bc81dbb1538","sourceUrl":"https://arxiv.org/abs/2611.12997","contentSha256":"05b30dba7686f280380887b03032dd695d566a77e570b52017ebaa206c9c4a6d","textHash":"867b41dd52bbf7e885db58e3c584098944a304ac93caf3b6fb3f2060070b6e70","observedAt":"2026-11-11T20:58:14.483Z"}},{"actor":{"name":"Liu, Peixi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Liu, Peixi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12997","locator":"head/meta[name=author|citation_author]"},"assertionId":"2d1203a376298200a7aa8bab65a44838e8a88607498d17adc3238bc81dbb1538:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:58:14.483Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"2d1203a376298200a7aa8bab65a44838e8a88607498d17adc3238bc81dbb1538","sourceUrl":"https://arxiv.org/abs/2611.12997","contentSha256":"05b30dba7686f280380887b03032dd695d566a77e570b52017ebaa206c9c4a6d","textHash":"867b41dd52bbf7e885db58e3c584098944a304ac93caf3b6fb3f2060070b6e70","observedAt":"2026-11-11T20:58:14.483Z"}},{"actor":{"name":"Jiao, Jianbo","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Jiao, Jianbo","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.12997","locator":"head/meta[name=author|citation_author]"},"assertionId":"2d1203a376298200a7aa8bab65a44838e8a88607498d17adc3238bc81dbb1538:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-11T20:58:14.483Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"2d1203a376298200a7aa8bab65a44838e8a88607498d17adc3238bc81dbb1538","sourceUrl":"https://arxiv.org/abs/2611.12997","contentSha256":"05b30dba7686f280380887b03032dd695d566a77e570b52017ebaa206c9c4a6d","textHash":"867b41dd52bbf7e885db58e3c584098944a304ac93caf3b6fb3f2060070b6e70","observedAt":"2026-11-11T20:58:14.483Z"}}],"participationCoverage":{"total":3,"returned":3,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.12997","snapshotId":"2d1203a376298200a7aa8bab65a44838e8a88607498d17adc3238bc81dbb1538","contentSha256":"05b30dba7686f280380887b03032dd695d566a77e570b52017ebaa206c9c4a6d","textHash":"867b41dd52bbf7e885db58e3c584098944a304ac93caf3b6fb3f2060070b6e70","observedAt":"2026-11-11T20:58:14.483Z","basis":"publisher-page-metadata","sourceLinks":43,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/56418bb94a53143cd01ef6cbecb892a9abb0eeb7d58300ea29df2777b1105055/source"}}}],"audits":[{"publicationId":"c111a40f-bbec-4157-8648-1d46e8ba36ca","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The paper and its claims are correctly attributed to Qiming Huang, Peixi Liu and Jianbo Jiao on arXiv. The theoretical and experimental assertions stay framed as the authors' claims. The paper is listed primarily under cs.CV and cross-listed to cs.AI, so the 'arXiv cs.AI' label is acceptable.","excerpt":"Authors:Qiming Huang, Peixi Liu, Jianbo Jiao"},{"dimension":"summary","status":"uncertain","explanation":"The summary is faithful, but central terms remain opaque to a general reader. It never explains plainly what 'attribution' means here: showing which parts of an image or text drove a model's output. 'Information bottleneck (IB)', 'proxy supervision' and 'posterior' are left unexplained. The reading also gives little evidence-supported sense of why the failure mode matters, such as explanations of vision-language models being misleading on noisy real-world data. Smallest repair: add a plain-language gloss of attribution and IB, or term-library tooltips, plus a conditional line on significance tied to the abstract.","excerpt":"Multimodal (e.g. vision-language, in this work) attribution aims to interpret the models without access to ground-truth explanatory supervision."},{"dimension":"date","status":"pass","explanation":"The originalPublished value of 2026-11-08 matches the v1 submission date on the arXiv page. It is not the later feed date of 2026-11-11.","excerpt":"[Submitted on 8 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim is supported by a verbatim abstract excerpt and kept as source-stated. The limitations correctly note that the abstract gives no benchmark details, results or reliability measure. The question is a genuine open issue rather than a leading claim, and no benefits or costs are invented.","excerpt":"by assuming equal reliability across all proxy supervision signals, they induce posterior over-contraction and forced explanations"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper, which is the most direct access to the source.","excerpt":null}],"createdAt":"2026-11-11T18:50:19.620Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}