A General Framework for Adversarial Examples with Objectives
This addresses the challenge of creating more practical and effective adversarial examples for applications such as security testing and face recognition, though it is incremental as it builds on existing adversarial example methods.
The authors tackled the problem of generating adversarial examples that satisfy additional real-world objectives beyond just similarity to original images, by proposing adversarial generative nets (AGNs) to train a generator for this purpose, resulting in improved robustness, inconspicuousness, and scalability for physical adversarial examples like eyeglass frames that fool face recognition.
Images perturbed subtly to be misclassified by neural networks, called adversarial examples, have emerged as a technically deep challenge and an important concern for several application domains. Most research on adversarial examples takes as its only constraint that the perturbed images are similar to the originals. However, real-world application of these ideas often requires the examples to satisfy additional objectives, which are typically enforced through custom modifications of the perturbation process. In this paper, we propose adversarial generative nets (AGNs), a general methodology to train a generator neural network to emit adversarial examples satisfying desired objectives. We demonstrate the ability of AGNs to accommodate a wide range of objectives, including imprecise ones difficult to model, in two application domains. In particular, we demonstrate physical adversarial examples---eyeglass frames designed to fool face recognition---with better robustness, inconspicuousness, and scalability than previous approaches, as well as a new attack to fool a handwritten-digit classifier.