Defending Adversarial Examples via DNN Bottleneck Reinforcement - Archive ouverte HAL Access content directly
Conference Papers Year : 2020

Defending Adversarial Examples via DNN Bottleneck Reinforcement

Wenqing Liu
  • Function : Author
  • PersonId : 1075447
Li Li
  • Function : Author
  • PersonId : 955781


This paper presents a DNN bottleneck reinforcement scheme to alleviate the vulnerability of Deep Neural Networks (DNN) against adversarial attacks. Typical DNN classifiers encode the input image into a compressed latent representation more suitable for inference. This information bottleneck makes a trade-off between the image-specific structure and class-specific information in an image. By reinforcing the former while maintaining the latter, any redundant information, be it adversarial or not, should be removed from the latent representation. Hence, this paper proposes to jointly train an auto-encoder (AE) sharing the same encoding weights with the visual classifier. In order to reinforce the information bottleneck, we introduce the multi-scale low-pass objective and multi-scale high-frequency communication for better frequency steering in the network. Unlike existing approaches, our scheme is the first reforming defense per se which keeps the classifier structure untouched without appending any pre-processing head and is trained with clean images only. Extensive experiments on MNIST, CIFAR-10 and ImageNet demonstrate the strong defense of our method against various adversarial attacks.
Fichier principal
Vignette du fichier
ACMMM_adversarial-3.pdf (1.94 Mo) Télécharger le fichier
Origin : Publisher files allowed on an open archive

Dates and versions

hal-02912189 , version 1 (20-11-2020)



Wenqing Liu, Miaojing Shi, Teddy Furon, Li Li. Defending Adversarial Examples via DNN Bottleneck Reinforcement. ACM Multimedia Conference 2020, Oct 2020, Seattle, United States. pp.1930-1938, ⟨10.1145/3394171.3413825⟩. ⟨hal-02912189⟩
519 View
229 Download



Gmail Facebook X LinkedIn More