Investigating Vulnerabilities Of Deep Neural Policies

Abstract

Reinforcement learning policies based on deep neural networks are vulnerable to imperceptible adversarial perturbations to their inputs, in much the same way as neural network image classifiers. Recent work has proposed several methods to improve the robustness of deep reinforcement learning agents to adversarial perturbations based on training in the presence of these imperceptible perturbations (i.e. adversarial training). In this paper, we study the effects of adversarial training on the neural policy learned by the agent. In particular, we follow two distinct parallel approaches to investigate the outcomes of adversarial training on deep neural policies based on worst-case distributional shift and feature sensitivity. For the first approach, we compare the Fourier spectrum of minimal perturbations computed for both adversarially trained and vanilla trained neural policies. Via experiments in the OpenAI Atari environments we show that minimal perturbations computed for adversarially

Stats

• citations0
• S2 citations—
• github stars0
• HF likes0
• heat score0.00
• arxiv keykorkmaz2021investigating

Related papers

• Minimalistic Attacks: How Little It Takes To Fool A Deep Reinforcement Learning Policy (2019)0.00
• Adversarial Policies: Attacking Deep Reinforcement Learning (2019)0.00
• Detecting Adversarial Attacks On Neural Network Policies With Visual Foresight (2017)0.00
• Real-time Adversarial Perturbations Against Deep Reinforcement Learning Policies: Attacks And Defenses (2021)0.00
• Improving Robustness Of Deep Reinforcement Learning Agents: Environment Attack Based On The Critic Network (2021)5.84
• Toward Evaluating Robustness Of Reinforcement Learning With Adversarial Policy (2023)4.52
• Attacking And Defending Deep Reinforcement Learning Policies (2022)0.00
• Defending Observation Attacks In Deep Reinforcement Learning Via Detection And Denoising (2022)0.00