Skip to main content

Defense through diverse directions

Publication ,  Conference
Bender, CM; Li, Y; Shi, Y; Reiter, MK; Oliva, JB
Published in: 37th International Conference on Machine Learning, ICML 2020
January 1, 2020

In this work we develop a novel Bayesian neural network methodology to achieve strong adversarial robustness without the need for online adversarial training. Unlike previous efforts in this direction, we do not rely solely on the stochasticity of network weights by minimizing the divergence between the learned parameter distribution and a prior. Instead, we additionally require that the model maintain some expected uncertainty with respect to all input covariates. We demonstrate that by encouraging the network to distribute evenly across inputs, the network becomes less susceptible to localized, brittle features which imparts a natural robustness to targeted perturbations. We show empirical robustness on several benchmark datasets.

Duke Scholars

Published In

37th International Conference on Machine Learning, ICML 2020

ISBN

9781713821120

Publication Date

January 1, 2020

Volume

PartF168147-1

Start / End Page

733 / 743
 

Citation

APA
Chicago
ICMJE
MLA
NLM
Bender, C. M., Li, Y., Shi, Y., Reiter, M. K., & Oliva, J. B. (2020). Defense through diverse directions. In 37th International Conference on Machine Learning, ICML 2020 (Vol. PartF168147-1, pp. 733–743).
Bender, C. M., Y. Li, Y. Shi, M. K. Reiter, and J. B. Oliva. “Defense through diverse directions.” In 37th International Conference on Machine Learning, ICML 2020, PartF168147-1:733–43, 2020.
Bender CM, Li Y, Shi Y, Reiter MK, Oliva JB. Defense through diverse directions. In: 37th International Conference on Machine Learning, ICML 2020. 2020. p. 733–43.
Bender, C. M., et al. “Defense through diverse directions.” 37th International Conference on Machine Learning, ICML 2020, vol. PartF168147-1, 2020, pp. 733–43.
Bender CM, Li Y, Shi Y, Reiter MK, Oliva JB. Defense through diverse directions. 37th International Conference on Machine Learning, ICML 2020. 2020. p. 733–743.

Published In

37th International Conference on Machine Learning, ICML 2020

ISBN

9781713821120

Publication Date

January 1, 2020

Volume

PartF168147-1

Start / End Page

733 / 743