In plain words: Cleaning an image before it reaches a classifier — smoothing out noise or rebuilding it from small patches — blocks attacks, and its randomness makes it hard for attackers to adapt. The best version stopped 90% of strong attacks where the attacker never sees the cleanup.
Abstract · Countering Adversarial Images using Input Transformations
This paper investigates strategies that defend against adversarial-example attacks on image-classification systems by transforming the inputs before feeding them to the system. Specifically, we study applying image transformations such as bit-depth reduction, JPEG compression, total variance minimization, and image quilting before feeding the image to a convolutional network classifier. Our experiments on ImageNet show that total variance minimization and image quilting are very effective defenses in practice, in particular, when the network is trained on transformed images. The strength of those defenses lies in their non-differentiable nature and their inherent randomness, which makes it difficult for an adversary to circumvent the defenses. Our best defense eliminates 60% of strong gray-box and 90% of strong black-box attacks by a variety of major attack methods
Chuan Guo, Mayank Rana, Moustapha Cisse, Laurens van der Maaten
arXiv:1711.00117 · cs.CV · submitted Oct 31, 2017 · updated Jan 25, 2018
abstract · pdf · html · 12 pages, 6 figures, submitted to ICLR 2018
