Adversarial Sample Detection Through Neural Network Transport Dynamics - Archive ouverte HAL Access content directly
Conference Papers Year : 2023

Adversarial Sample Detection Through Neural Network Transport Dynamics

Abstract

We propose a detector of adversarial samples that is based on the view of neural networks as discrete dynamic systems. The detector tells clean inputs from abnormal ones by comparing the discrete vector fields they follow through the layers. We also show that regularizing this vector field during training makes the network more regular on the data distribution's support, thus making the activations of clean inputs more distinguishable from those of abnormal ones. Experimentally, we compare our detector favorably to other detectors on seen and unseen attacks, and show that the regularization of the network's dynamics improves the performance of adversarial detectors that use the internal embeddings as inputs, while also improving test accuracy.
Fichier principal
Vignette du fichier
main.pdf (679.54 Ko) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

hal-04120861 , version 1 (07-06-2023)

Identifiers

  • HAL Id : hal-04120861 , version 1

Cite

Skander Karkar, Patrick Gallinari, Alain Rakotomamonjy. Adversarial Sample Detection Through Neural Network Transport Dynamics. European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD 2023), Sep 2023, Torino, Italy. ⟨hal-04120861⟩
46 View
75 Download

Share

Gmail Mastodon Facebook X LinkedIn More