Open Peer Review. Open Publishing. Open Access. Open Discussion. Open Directory. Open Recommendations. Open API. Open Source.
Countering Adversarial Images using Input Transformations
Nov 03, 2017 (modified: Nov 03, 2017)ICLR 2018 Conference Blind Submissionreaders: everyoneShow Bibtex
Abstract: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 white-box and 90% of strong black-box attacks by a variety of major attack methods.
TL;DR:We apply a model-agnostic defense strategy against adversarial examples and achieve 60% white-box accuracy and 90% black-box accuracy against major attack algorithms.
Keywords:adversarial example, machine learning security, computer vision, image classification
Enter your feedback below and we'll get back to you as soon as possible.