Matches in SemOpenAlex for { <https://semopenalex.org/work/W3158164920> ?p ?o ?g. }
- W3158164920 abstract "An important pillar for safe machine learning (ML) is the systematic mitigation of weaknesses in neural networks to afford their deployment in critical applications. An ubiquitous class of safety risks are learned shortcuts, i.e. spurious correlations a network exploits for its decisions that have no semantic connection to the actual task. Networks relying on such shortcuts bear the risk of not generalizing well to unseen inputs. Explainability methods help to uncover such network vulnerabilities. However, many of these techniques are not directly applicable if access to the network is constrained, in so-called black-box setups. These setups are prevalent when using third-party ML components. To address this constraint, we present an approach to detect learned shortcuts using an interpretable-by-design network as a proxy to the black-box model of interest. Leveraging the proxy's guarantees on introspection we automatically extract candidates for learned shortcuts. Their transferability to the black box is validated in a systematic fashion. Concretely, as proxy model we choose a BagNet, which bases its decisions purely on local image patches. We demonstrate on the autonomous driving dataset A2D2 that extracted patch shortcuts significantly influence the black box model. By efficiently identifying such patch-based vulnerabilities, we contribute to safer ML models." @default.
- W3158164920 created "2021-05-10" @default.
- W3158164920 creator A5011846200 @default.
- W3158164920 creator A5020225503 @default.
- W3158164920 creator A5033826340 @default.
- W3158164920 creator A5038925389 @default.
- W3158164920 creator A5062749833 @default.
- W3158164920 date "2021-04-22" @default.
- W3158164920 modified "2023-09-28" @default.
- W3158164920 title "Patch Shortcuts: Interpretable Proxy Models Efficiently Find Black-Box Vulnerabilities" @default.
- W3158164920 cites W1522301498 @default.
- W3158164920 cites W1799366690 @default.
- W3158164920 cites W1821462560 @default.
- W3158164920 cites W1999360130 @default.
- W3158164920 cites W2180612164 @default.
- W3158164920 cites W2194775991 @default.
- W3158164920 cites W2282821441 @default.
- W3158164920 cites W2408141691 @default.
- W3158164920 cites W2603766943 @default.
- W3158164920 cites W2746314669 @default.
- W3158164920 cites W2769421449 @default.
- W3158164920 cites W2798302089 @default.
- W3158164920 cites W2809136100 @default.
- W3158164920 cites W2885106262 @default.
- W3158164920 cites W2888394138 @default.
- W3158164920 cites W2890054895 @default.
- W3158164920 cites W2903703378 @default.
- W3158164920 cites W2942630857 @default.
- W3158164920 cites W2945526826 @default.
- W3158164920 cites W2949736877 @default.
- W3158164920 cites W2953610242 @default.
- W3158164920 cites W2961301154 @default.
- W3158164920 cites W2962680264 @default.
- W3158164920 cites W2962748759 @default.
- W3158164920 cites W2962768284 @default.
- W3158164920 cites W2962772482 @default.
- W3158164920 cites W2962858109 @default.
- W3158164920 cites W2962949867 @default.
- W3158164920 cites W2963540976 @default.
- W3158164920 cites W2963726920 @default.
- W3158164920 cites W2963847595 @default.
- W3158164920 cites W2964077693 @default.
- W3158164920 cites W2964222566 @default.
- W3158164920 cites W2970030610 @default.
- W3158164920 cites W2970222187 @default.
- W3158164920 cites W2971048680 @default.
- W3158164920 cites W2990289029 @default.
- W3158164920 cites W2995404272 @default.
- W3158164920 cites W3016404858 @default.
- W3158164920 cites W3035661013 @default.
- W3158164920 cites W3085109610 @default.
- W3158164920 cites W3100149641 @default.
- W3158164920 cites W3100511085 @default.
- W3158164920 cites W3104310207 @default.
- W3158164920 cites W3107235539 @default.
- W3158164920 cites W3108072218 @default.
- W3158164920 cites W3110564906 @default.
- W3158164920 cites W3151681879 @default.
- W3158164920 hasPublicationYear "2021" @default.
- W3158164920 type Work @default.
- W3158164920 sameAs 3158164920 @default.
- W3158164920 citedByCount "0" @default.
- W3158164920 crossrefType "posted-content" @default.
- W3158164920 hasAuthorship W3158164920A5011846200 @default.
- W3158164920 hasAuthorship W3158164920A5020225503 @default.
- W3158164920 hasAuthorship W3158164920A5033826340 @default.
- W3158164920 hasAuthorship W3158164920A5038925389 @default.
- W3158164920 hasAuthorship W3158164920A5062749833 @default.
- W3158164920 hasConcept C119857082 @default.
- W3158164920 hasConcept C124101348 @default.
- W3158164920 hasConcept C154945302 @default.
- W3158164920 hasConcept C165696696 @default.
- W3158164920 hasConcept C2780148112 @default.
- W3158164920 hasConcept C2984842247 @default.
- W3158164920 hasConcept C38652104 @default.
- W3158164920 hasConcept C41008148 @default.
- W3158164920 hasConcept C50644808 @default.
- W3158164920 hasConcept C94966114 @default.
- W3158164920 hasConcept C97256817 @default.
- W3158164920 hasConceptScore W3158164920C119857082 @default.
- W3158164920 hasConceptScore W3158164920C124101348 @default.
- W3158164920 hasConceptScore W3158164920C154945302 @default.
- W3158164920 hasConceptScore W3158164920C165696696 @default.
- W3158164920 hasConceptScore W3158164920C2780148112 @default.
- W3158164920 hasConceptScore W3158164920C2984842247 @default.
- W3158164920 hasConceptScore W3158164920C38652104 @default.
- W3158164920 hasConceptScore W3158164920C41008148 @default.
- W3158164920 hasConceptScore W3158164920C50644808 @default.
- W3158164920 hasConceptScore W3158164920C94966114 @default.
- W3158164920 hasConceptScore W3158164920C97256817 @default.
- W3158164920 hasOpenAccess W3158164920 @default.
- W3158164920 hasRelatedWork W2142895442 @default.
- W3158164920 hasRelatedWork W2216964862 @default.
- W3158164920 hasRelatedWork W2243991213 @default.
- W3158164920 hasRelatedWork W2279943773 @default.
- W3158164920 hasRelatedWork W2397821426 @default.
- W3158164920 hasRelatedWork W2513989232 @default.
- W3158164920 hasRelatedWork W2911399349 @default.
- W3158164920 hasRelatedWork W2965367174 @default.
- W3158164920 hasRelatedWork W2967753664 @default.