Matches in SemOpenAlex for { <https://semopenalex.org/work/W2773300778> ?p ?o ?g. }
Showing items 1 to 100 of
100
with 100 items per page.
- W2773300778 abstract "Recent studies have revealed the vulnerability of deep neural networks: A small adversarial perturbation that is imperceptible to human can easily make a well-trained deep neural network misclassify. This makes it unsafe to apply neural networks in security-critical applications. In this paper, we propose a new defense algorithm called Random Self-Ensemble (RSE) by combining two important concepts: {bf randomness} and {bf ensemble}. To protect a targeted model, RSE adds random noise layers to the neural network to prevent the strong gradient-based attacks, and ensembles the prediction over random noises to stabilize the performance. We show that our algorithm is equivalent to ensemble an infinite number of noisy models $f_epsilon$ without any additional memory overhead, and the proposed training procedure based on noisy stochastic gradient descent can ensure the ensemble model has a good predictive capability. Our algorithm significantly outperforms previous defense techniques on real data sets. For instance, on CIFAR-10 with VGG network (which has 92% accuracy without any attack), under the strong C&W attack within a certain distortion tolerance, the accuracy of unprotected model drops to less than 10%, the best previous defense technique has $48%$ accuracy, while our method still has $86%$ prediction accuracy under the same level of attack. Finally, our method is simple and easy to integrate into any neural network." @default.
- W2773300778 created "2017-12-22" @default.
- W2773300778 creator A5000534132 @default.
- W2773300778 creator A5005915688 @default.
- W2773300778 creator A5010841999 @default.
- W2773300778 creator A5024120345 @default.
- W2773300778 date "2017-12-02" @default.
- W2773300778 modified "2023-09-23" @default.
- W2773300778 title "Towards Robust Neural Networks via Random Self-ensemble" @default.
- W2773300778 cites W1514535095 @default.
- W2773300778 cites W2097117768 @default.
- W2773300778 cites W2194775991 @default.
- W2773300778 cites W2230740169 @default.
- W2773300778 cites W2549139847 @default.
- W2773300778 cites W2552767274 @default.
- W2773300778 cites W2590523583 @default.
- W2773300778 cites W2619479788 @default.
- W2773300778 cites W2755956102 @default.
- W2773300778 cites W2765233338 @default.
- W2773300778 cites W2765384636 @default.
- W2773300778 cites W2786118190 @default.
- W2773300778 cites W2786977288 @default.
- W2773300778 cites W2787708942 @default.
- W2773300778 cites W2787733970 @default.
- W2773300778 cites W2798302089 @default.
- W2773300778 cites W2949311987 @default.
- W2773300778 cites W2951807304 @default.
- W2773300778 cites W2962835968 @default.
- W2773300778 cites W2963158386 @default.
- W2773300778 cites W2963207607 @default.
- W2773300778 cites W2963334011 @default.
- W2773300778 cites W2963448658 @default.
- W2773300778 cites W2963612069 @default.
- W2773300778 cites W2963626858 @default.
- W2773300778 cites W2963744840 @default.
- W2773300778 cites W2963857521 @default.
- W2773300778 cites W2964082701 @default.
- W2773300778 cites W2964153729 @default.
- W2773300778 cites W2964197269 @default.
- W2773300778 cites W2964253222 @default.
- W2773300778 doi "https://doi.org/10.48550/arxiv.1712.00673" @default.
- W2773300778 hasPublicationYear "2017" @default.
- W2773300778 type Work @default.
- W2773300778 sameAs 2773300778 @default.
- W2773300778 citedByCount "17" @default.
- W2773300778 countsByYear W27733007782018 @default.
- W2773300778 countsByYear W27733007782019 @default.
- W2773300778 countsByYear W27733007782020 @default.
- W2773300778 countsByYear W27733007782021 @default.
- W2773300778 crossrefType "posted-content" @default.
- W2773300778 hasAuthorship W2773300778A5000534132 @default.
- W2773300778 hasAuthorship W2773300778A5005915688 @default.
- W2773300778 hasAuthorship W2773300778A5010841999 @default.
- W2773300778 hasAuthorship W2773300778A5024120345 @default.
- W2773300778 hasBestOaLocation W27733007781 @default.
- W2773300778 hasConcept C105795698 @default.
- W2773300778 hasConcept C108583219 @default.
- W2773300778 hasConcept C11413529 @default.
- W2773300778 hasConcept C119857082 @default.
- W2773300778 hasConcept C119898033 @default.
- W2773300778 hasConcept C125112378 @default.
- W2773300778 hasConcept C154945302 @default.
- W2773300778 hasConcept C2984842247 @default.
- W2773300778 hasConcept C2986577269 @default.
- W2773300778 hasConcept C33923547 @default.
- W2773300778 hasConcept C38652104 @default.
- W2773300778 hasConcept C41008148 @default.
- W2773300778 hasConcept C50644808 @default.
- W2773300778 hasConcept C95713431 @default.
- W2773300778 hasConceptScore W2773300778C105795698 @default.
- W2773300778 hasConceptScore W2773300778C108583219 @default.
- W2773300778 hasConceptScore W2773300778C11413529 @default.
- W2773300778 hasConceptScore W2773300778C119857082 @default.
- W2773300778 hasConceptScore W2773300778C119898033 @default.
- W2773300778 hasConceptScore W2773300778C125112378 @default.
- W2773300778 hasConceptScore W2773300778C154945302 @default.
- W2773300778 hasConceptScore W2773300778C2984842247 @default.
- W2773300778 hasConceptScore W2773300778C2986577269 @default.
- W2773300778 hasConceptScore W2773300778C33923547 @default.
- W2773300778 hasConceptScore W2773300778C38652104 @default.
- W2773300778 hasConceptScore W2773300778C41008148 @default.
- W2773300778 hasConceptScore W2773300778C50644808 @default.
- W2773300778 hasConceptScore W2773300778C95713431 @default.
- W2773300778 hasLocation W27733007781 @default.
- W2773300778 hasOpenAccess W2773300778 @default.
- W2773300778 hasPrimaryLocation W27733007781 @default.
- W2773300778 hasRelatedWork W2791691546 @default.
- W2773300778 hasRelatedWork W2950066684 @default.
- W2773300778 hasRelatedWork W3082895349 @default.
- W2773300778 hasRelatedWork W3136979370 @default.
- W2773300778 hasRelatedWork W3179488938 @default.
- W2773300778 hasRelatedWork W4288853838 @default.
- W2773300778 hasRelatedWork W4298388782 @default.
- W2773300778 hasRelatedWork W4312831135 @default.
- W2773300778 hasRelatedWork W4317565044 @default.
- W2773300778 hasRelatedWork W4318677156 @default.
- W2773300778 isParatext "false" @default.
- W2773300778 isRetracted "false" @default.
- W2773300778 magId "2773300778" @default.
- W2773300778 workType "article" @default.