HAL will be down for maintenance from Friday, June 10 at 4pm through Monday, June 13 at 9am. More information
Skip to Main content Skip to Navigation
Conference papers

Impact of reverberation through deep neural networks on adversarial perturbations

Abstract : The vulnerability of Deep Neural Network (DNN) models to maliciously crafted adversarial perturbations is a critical topic considering their ongoing large-scale deployment. In this work, we explore an interesting phenomenon that occurs when an image is reinjected multiple times into a DNN, according to a procedure (called reverberation) that has been first proposed in cognitive psychology to avoid the catastrophic forgetting issue, through its impact on adversarial perturbations. We describe reverberation in vanilla autoencoders and propose a new reverberant architecture combining a classifier and an autoencoder that allows the joint observation of the logits and reconstructed images. We experimentally measure the impact of reverberation on adversarial perturbations placing ourselves in a scenario of adversarial example detection. The results show that clean and adversarial examples even with small levels of perturbation behave very differently throughout reverberation. While computationally efficient (reverberation is only based on inferences), our approach yields promising results for adversarial examples detection, consistent across datasets, adversarial attacks and DNN architectures.
Document type :
Conference papers
Complete list of metadata

https://hal-cea.archives-ouvertes.fr/cea-03605422
Contributor : Contributeur Map Cea Connect in order to contact the contributor
Submitted on : Friday, March 11, 2022 - 9:00:08 AM
Last modification on : Sunday, March 13, 2022 - 3:26:18 AM

File

ICMLA2021_paper155_camera-read...
Files produced by the author(s)

Identifiers

  • HAL Id : cea-03605422, version 1

Citation

Romain Cohendet, Miguel Angel Solinas, Rémi Bernhard, Marina Reyboz, Pierre-Alain Moellic, et al.. Impact of reverberation through deep neural networks on adversarial perturbations. ICMLA 2021 - 20th IEEE International Cconference on Machine Learning and Applications, Dec 2021, Pasadena (Virtual event), United States. ⟨cea-03605422⟩

Share

Metrics

Record views

9

Files downloads

8