Systematic Vulnerability Audit of Post-Hoc XAI under Common Corruptions: A Factor Analysis Across Vision Benchmarks

Systematic Vulnerability Audit of Post-Hoc XAI under Common Corruptions: A Factor Analysis Across Vision Benchmarks | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Systematic Vulnerability Audit of Post-Hoc XAI under Common Corruptions: A Factor Analysis Across Vision Benchmarks Minyeong Kim This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9501010/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Post-hoc attribution methods are widely deployed to explain deep vision classifiers, yet no systematic evaluation protocol exists for the corrupted-input regime. This paper introduces the first such protocol for attribution stability under distribution shift, validated through a factorial audit of five attribution methods (Integrated Gradients, Grad-CAM, SmoothGrad, GradientSHAP, LIME) under fifteen corruptions at five severity levels on CIFAR-10-C and CIFAR-100-C, across two architecturally distinct classifiers (ResNet-50, ViT-B/16), yielding 760,000 clean–corrupted pairs and five stability metrics. Four findings validate the protocol. First, attribution stability declines monotonically with corruption severity for all methods, with 12 to 13 of 15 corruption types reaching significance under Benjamini–Hochberg correction and photometric corruptions exhibiting the steepest degradation (brightness Spearman: 0.87 at severity 1 to 0.31 at severity 5). Second, degradation magnitude depends dramatically on the method: SmoothGrad retains Spearman 0.91 for brightness at severity 3 while LIME falls to 0.04, a twenty-fold gap. Third, the resolution-fair ranking (SmoothGrad, IG, GradientSHAP) is consistent across 96% of cells (144/150), with method η2 > 0.84 dwarfing architecture η2 < 0.02. Fourth, the architecture-by-method interaction is non-significant on CIFAR-10 (p = 0.71) but significant on CIFAR- 100 (p = 0.035), showing task-complexity modulation of the model-agnostic claim. We contribute three artifacts: (i) the protocol as an evaluation standard, (ii) an empirical method characterisation, and (iii) a deployment-ready ranking; we show the model-agnostic assumption requires re-examination as classification complexity increases. Explainable Artificial Intelligence Attribution Stability Distribution Shift Common Corruptions Vision Transformer Model-Agnostic Explanation Full Text Additional Declarations No competing interests reported. Supplementary Files 03informationsheet.pdf Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9501010","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":628611651,"identity":"61a8e16d-4830-47b2-846f-497c9b611028","order_by":0,"name":"Minyeong Kim","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAsElEQVRIiWNgGAWjYDACZgaGAx8qbEjTwnhwxpk0Eu05zNt2mAT1Bsd5DxzmOXM+j39G8rEPDDU20YS1HOZLODin4naxxI205BkMx9JyGwhpMTvMY3DgzZnbiQ1nzhgzMDYcJlILb9u5xPlnzn8mXstB3rYDiRuO9zATp8UepGXGmeTEjcfbjBkSiPGLZP8Z4w8fKuwS5x1mfszwocaGsBZUkECa8lEwCkbBKBgFuAAAvgpFdnbHTRQAAAAASUVORK5CYII=","orcid":"","institution":"","correspondingAuthor":true,"prefix":"","firstName":"Minyeong","middleName":"","lastName":"Kim","suffix":""}],"badges":[],"createdAt":"2026-04-23 01:23:31","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9501010/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9501010/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107730412,"identity":"c6a2d6ff-700f-4c3f-be97-fc6dfe910da1","added_by":"auto","created_at":"2026-04-24 12:56:41","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":769961,"visible":true,"origin":"","legend":"","description":"","filename":"mljsource.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9501010/v1_covered_9cf3db37-982d-41d1-bb13-5d6a899098a0.pdf"},{"id":107684102,"identity":"af407d42-a46b-43ec-b59c-4eb6e6a87463","added_by":"auto","created_at":"2026-04-24 03:44:19","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":134062,"visible":true,"origin":"","legend":"","description":"","filename":"03informationsheet.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9501010/v1/9fa85dd3a9d25336e8b92311.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Systematic Vulnerability Audit of Post-Hoc XAI under Common Corruptions: A Factor Analysis Across Vision Benchmarks","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Explainable Artificial Intelligence, Attribution Stability, Distribution Shift, Common Corruptions, Vision Transformer, Model-Agnostic Explanation","lastPublishedDoi":"10.21203/rs.3.rs-9501010/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9501010/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003ePost-hoc attribution methods are widely deployed to explain deep vision classifiers, yet no systematic evaluation protocol exists for the corrupted-input regime. This paper introduces the first such protocol for attribution stability under distribution shift, validated through a factorial audit of five attribution methods (Integrated Gradients, Grad-CAM, SmoothGrad, GradientSHAP, LIME) under fifteen corruptions at five severity levels on CIFAR-10-C and CIFAR-100-C, across two architecturally distinct classifiers (ResNet-50, ViT-B/16), yielding 760,000 clean–corrupted pairs and five stability metrics. Four findings validate the protocol. First, attribution stability declines monotonically with corruption severity for all methods, with 12 to 13 of 15 corruption types reaching significance under Benjamini–Hochberg correction and photometric corruptions exhibiting the steepest degradation (brightness Spearman: 0.87 at severity 1 to 0.31 at severity 5). Second, degradation magnitude depends dramatically on the method: SmoothGrad retains Spearman 0.91 for brightness at severity 3 while LIME falls to 0.04, a twenty-fold gap. Third, the resolution-fair ranking (SmoothGrad, IG, GradientSHAP) is consistent across 96% of cells (144/150), with method η2 \u0026gt; 0.84 dwarfing architecture η2 \u0026lt; 0.02. Fourth, the architecture-by-method interaction is non-significant on CIFAR-10 (p = 0.71) but significant on CIFAR- 100 (p = 0.035), showing task-complexity modulation of the model-agnostic claim. We contribute three artifacts: (i) the protocol as an evaluation standard, (ii) an empirical method characterisation, and (iii) a deployment-ready ranking; we show the model-agnostic assumption requires re-examination as classification complexity increases.\u003c/p\u003e","manuscriptTitle":"Systematic Vulnerability Audit of Post-Hoc XAI under Common Corruptions: A Factor Analysis Across Vision Benchmarks","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-24 03:44:16","doi":"10.21203/rs.3.rs-9501010/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"fbc92cf3-4d3a-4dda-b4de-a5e72fb51659","owner":[],"postedDate":"April 24th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-24T12:55:15+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-24 03:44:16","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9501010","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9501010","identity":"rs-9501010","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}