Adversarial Detection and Correction by Matching Prediction Distributions

Abstract

We present a novel adversarial detection and correction method for machine learning classifiers.The detector consists of an autoencoder trained with a custom loss function based on the Kullback-Leibler divergence between the classifier predictions on the original and reconstructed instances.The method is unsupervised, easy to train and does not require any knowledge about the underlying attack. The detector almost completely neutralises powerful attacks like Carlini-Wagner or SLIDE on MNIST and Fashion-MNIST, and remains very effective on CIFAR-10 when the attack is granted full access to the classification model but not the defence. We show that our method is still able to detect the adversarial examples in the case of a white-box attack where the attacker has full knowledge of both the model and the defence and investigate the robustness of the attack. The method is very flexible and can also be used to detect common data corruptions and perturbations which negatively impact the model performance. We illustrate this capability on the CIFAR-10-C dataset.

13 pages, 16 figures. For an open source implementation of the algorithm, see https://github.com/SeldonIO/alibi-detect

References in corpus (8)

• Distilling the Knowledge in a Neural Network
• Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
• Intriguing properties of neural networks
• Failing Loudly: An Empirical Study of Methods for Detecting Dataset Shift
• Adversarial Training and Robustness for Multiple Perturbations
• EAD: Elastic-Net Attacks to Deep Neural Networks via Adversarial Examples
• Defense-VAE: A Fast and Accurate Defense against Adversarial Attacks
• Purifying Adversarial Perturbation with Adversarially Trained Auto-encoders

Cited by in corpus (2)

• Monitoring and explainability of models in production
• Two Souls in an Adversarial Image: Towards Universal Adversarial Example Detection using Multi-view Inconsistency