Online Robustness Training for Deep Reinforcement Learning

Marc Fischer; Matthew Mirman; Steven Stalder; Martin Vechev

Online Robustness Training for Deep Reinforcement Learning

Machine Learning 2019-11-25 v3 Machine Learning

Authors: Marc Fischer , Matthew Mirman , Steven Stalder , Martin Vechev

Abstract

In deep reinforcement learning (RL), adversarial attacks can trick an agent into unwanted states and disrupt training. We propose a system called Robust Student-DQN (RS-DQN), which permits online robustness training alongside Q networks, while preserving competitive performance. We show that RS-DQN can be combined with (i) state-of-the-art adversarial training and (ii) provably robust training to obtain an agent that is resilient to strong attacks during training and evaluation.

Keywords

adversarial training adversarial examples reinforcement learning

Cite

@article{arxiv.1911.00887,
  title  = {Online Robustness Training for Deep Reinforcement Learning},
  author = {Marc Fischer and Matthew Mirman and Steven Stalder and Martin Vechev},
  journal= {arXiv preprint arXiv:1911.00887},
  year   = {2019}
}

Related papers

View all related →

Artificial Intelligence · Computer Science

Whatever Does Not Kill Deep Reinforcement Learning, Makes It Stronger