Advances in adversarial attacks and defenses in computer vision: A survey

Deep Learning (DL) is the most widely used tool in the contemporary field of\ncomputer vision. Its ability to accurately solve complex problems is employed\nin vision research to learn deep neural models for a variety of tasks,\nincluding security critical applications. However, it is now known that DL is\nvulnerable to adversarial attacks that can manipulate its predictions by\nintroducing visually imperceptible perturbations in images and videos. Since\nthe discovery of this phenomenon in 2013~[1], it has attracted significant\nattention of researchers from multiple sub-fields of machine intelligence. In\n[2], we reviewed the contributions made by the computer vision community in\nadversarial attacks on deep learning (and their defenses) until the advent of\nyear 2018. Many of those contributions have inspired new directions in this\narea, which has matured significantly since witnessing the first generation\nmethods. Hence, as a legacy sequel of [2], this literature review focuses on\nthe advances in this area since 2018. To ensure authenticity, we mainly\nconsider peer-reviewed contributions published in the prestigious sources of\ncomputer vision and machine learning research. Besides a comprehensive\nliterature review, the article also provides concise definitions of technical\nterminologies for non-experts in this domain. Finally, this article discusses\nchallenges and future outlook of this direction based on the literature\nreviewed herein and [2].\n