Optimism in the Face of Adversity: Understanding and Improving Deep Learning through Adversarial Robustness

Driven by massive amounts of data and important advances in computational\nresources, new deep learning systems have achieved outstanding results in a\nlarge spectrum of applications. Nevertheless, our current theoretical\nunderstanding on the mathematical foundations of deep learning lags far behind\nits empirical success. Towards solving the vulnerability of neural networks,\nhowever, the field of adversarial robustness has recently become one of the\nmain sources of explanations of our deep models. In this article, we provide an\nin-depth review of the field of adversarial robustness in deep learning, and\ngive a self-contained introduction to its main notions. But, in contrast to the\nmainstream pessimistic perspective of adversarial robustness, we focus on the\nmain positive aspects that it entails. We highlight the intuitive connection\nbetween adversarial examples and the geometry of deep neural networks, and\neventually explore how the geometric study of adversarial examples can serve as\na powerful tool to understand deep learning. Furthermore, we demonstrate the\nbroad applicability of adversarial robustness, providing an overview of the\nmain emerging applications of adversarial robustness beyond security. The goal\nof this article is to provide readers with a set of new perspectives to\nunderstand deep learning, and to supply them with intuitive tools and insights\non how to use adversarial robustness to improve it.\n