Challenges and Countermeasures for Adversarial Attacks on Deep Reinforcement Learning

  • 2021-09-08 07:46:42
  • Inaam Ilahi, Muhammad Usama, Junaid Qadir, Muhammad Umar Janjua, Ala Al-Fuqaha, Dinh Thai Hoang, Dusit Niyato
  • 0


Deep Reinforcement Learning (DRL) has numerous applications in the real worldthanks to its outstanding ability in quickly adapting to the surroundingenvironments. Despite its great advantages, DRL is susceptible to adversarialattacks, which precludes its use in real-life critical systems and applications(e.g., smart grids, traffic controls, and autonomous vehicles) unless itsvulnerabilities are addressed and mitigated. Thus, this paper provides acomprehensive survey that discusses emerging attacks in DRL-based systems andthe potential countermeasures to defend against these attacks. We first coversome fundamental backgrounds about DRL and present emerging adversarial attackson machine learning techniques. We then investigate more details of thevulnerabilities that the adversary can exploit to attack DRL along with thestate-of-the-art countermeasures to prevent such attacks. Finally, we highlightopen issues and research challenges for developing solutions to deal withattacks for DRL-based intelligent systems.


Quick Read (beta)

loading the full paper ...