Minimalistic attacks : how little it takes to fool deep reinforcement learning policies

• Main Authors: Qu, Xinghua, Sun, Zhu, Ong, Yew-Soon, Gupta, Abhishek, Wei, Pengfei
• Other Authors: School of Computer Science and Engineering
• Format: Article
• Language: English
• Published: 2021
• Subjects: Engineering::Computer science and engineering; Reinforcement Learning; Adversarial Attacks
• Online Access: https://hdl.handle.net/10356/153700
• Institution: Nanyang Technological University

Recent studies have revealed that neural network-based policies can be easily fooled by adversarial examples. However, while most prior works analyze the effects of perturbing every pixel of every frame assuming white-box policy access, in this paper we take a more restrictive view towards adversary...