Matches in SemOpenAlex for { <https://semopenalex.org/work/W3120955712> ?p ?o ?g. }
Showing items 1 to 59 of
59
with 100 items per page.
- W3120955712 abstract "Neural networks are known to be vulnerable to adversarial attacks - small, imperceptible perturbations that cause the network to misclassify an input. A recent line of work attempts to explain this behavior by positing the existence of non-robust features - well-generalizing but brittle features present in the data distribution that are learned by the network and can be perturbed to cause misclassification. In this paper, we look at the dynamics of neural network training through the perspective of robust and non-robust features. We find that there are two very distinct pathways that neural network training can follow, depending on the hyperparameters used. In the first pathway, the network initially learns only predictive, robust features and weakly predictive non-robust features, and subsequently learns predictive, non-robust features. On the other hand, a network trained via the second pathway eschews predictive non-robust features altogether, and rapidly overfits the training data. We provide strong empirical evidence to corroborate this hypothesis, as well as theoretical analysis in a simplified setting. Key to our analysis is a better understanding of the relationship between predictive non-robust features and adversarial transferability. We present our findings in light of other recent results on the evolution of inductive biases learned by neural networks over the course of training. Finally, we digress to show that rather than being quirks of the data distribution, predictive non-robust features might actually occur across datasets with different distributions drawn from independent sources, indicating that they perhaps possess some meaning in terms of human semantics." @default.
- W3120955712 created "2021-01-18" @default.
- W3120955712 creator A5064541855 @default.
- W3120955712 date "2021-05-04" @default.
- W3120955712 modified "2023-09-26" @default.
- W3120955712 title "SGD on Neural Networks learns Robust Features before Non-Robust" @default.
- W3120955712 hasPublicationYear "2021" @default.
- W3120955712 type Work @default.
- W3120955712 sameAs 3120955712 @default.
- W3120955712 citedByCount "0" @default.
- W3120955712 crossrefType "journal-article" @default.
- W3120955712 hasAuthorship W3120955712A5064541855 @default.
- W3120955712 hasConcept C119857082 @default.
- W3120955712 hasConcept C121332964 @default.
- W3120955712 hasConcept C154945302 @default.
- W3120955712 hasConcept C2984842247 @default.
- W3120955712 hasConcept C37736160 @default.
- W3120955712 hasConcept C41008148 @default.
- W3120955712 hasConcept C50644808 @default.
- W3120955712 hasConcept C61797465 @default.
- W3120955712 hasConcept C62520636 @default.
- W3120955712 hasConcept C8642999 @default.
- W3120955712 hasConceptScore W3120955712C119857082 @default.
- W3120955712 hasConceptScore W3120955712C121332964 @default.
- W3120955712 hasConceptScore W3120955712C154945302 @default.
- W3120955712 hasConceptScore W3120955712C2984842247 @default.
- W3120955712 hasConceptScore W3120955712C37736160 @default.
- W3120955712 hasConceptScore W3120955712C41008148 @default.
- W3120955712 hasConceptScore W3120955712C50644808 @default.
- W3120955712 hasConceptScore W3120955712C61797465 @default.
- W3120955712 hasConceptScore W3120955712C62520636 @default.
- W3120955712 hasConceptScore W3120955712C8642999 @default.
- W3120955712 hasLocation W31209557121 @default.
- W3120955712 hasOpenAccess W3120955712 @default.
- W3120955712 hasPrimaryLocation W31209557121 @default.
- W3120955712 hasRelatedWork W2401467808 @default.
- W3120955712 hasRelatedWork W2897355816 @default.
- W3120955712 hasRelatedWork W2907518016 @default.
- W3120955712 hasRelatedWork W2944864651 @default.
- W3120955712 hasRelatedWork W2972561187 @default.
- W3120955712 hasRelatedWork W2981077337 @default.
- W3120955712 hasRelatedWork W2993049325 @default.
- W3120955712 hasRelatedWork W2997223761 @default.
- W3120955712 hasRelatedWork W3006227479 @default.
- W3120955712 hasRelatedWork W3041269312 @default.
- W3120955712 hasRelatedWork W3043393327 @default.
- W3120955712 hasRelatedWork W3088536850 @default.
- W3120955712 hasRelatedWork W3092696781 @default.
- W3120955712 hasRelatedWork W3099180329 @default.
- W3120955712 hasRelatedWork W3129273212 @default.
- W3120955712 hasRelatedWork W3134761266 @default.
- W3120955712 hasRelatedWork W3167454922 @default.
- W3120955712 hasRelatedWork W3177329815 @default.
- W3120955712 hasRelatedWork W3199086051 @default.
- W3120955712 hasRelatedWork W3208923398 @default.
- W3120955712 isParatext "false" @default.
- W3120955712 isRetracted "false" @default.
- W3120955712 magId "3120955712" @default.
- W3120955712 workType "article" @default.