Regret-based defense in adversarial reinforcement learning

• Main Authors: BELAIRE, Roman, VARAKANTHAM, Pradeep, NGUYEN, Thanh Hong, LO, David
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2024
• Subjects: Robust Reinforcement Learning; Adversarial Robustness; Regret; Software Engineering
• Online Access: https://ink.library.smu.edu.sg/sis_research/9243; https://ink.library.smu.edu.sg/context/sis_research/article/10243/viewcontent/p2633.pdf
• Institution: Singapore Management University

Deep Reinforcement Learning (DRL) policies are vulnerable to adversarial noise in observations, which can have disastrous consequences in safety-critical environments. For instance, a self-driving car receiving adversarially perturbed sensory observations about traffic signs (e.g., a stop sign physi...