Adaptive Adversarial Training for Meta Reinforcement Learning

Shiqi Chen; Zhengyu Chen; Donglin Wang

Adaptive Adversarial Training for Meta Reinforcement Learning

Machine Learning 2021-04-28 v1 Artificial Intelligence

Authors: Shiqi Chen , Zhengyu Chen , Donglin Wang

Abstract

Meta Reinforcement Learning (MRL) enables an agent to learn from a limited number of past trajectories and extrapolate to a new task. In this paper, we attempt to improve the robustness of MRL. We build upon model-agnostic meta-learning (MAML) and propose a novel method to generate adversarial samples for MRL by using Generative Adversarial Network (GAN). That allows us to enhance the robustness of MRL to adversal attacks by leveraging these attacks during meta training process.

Keywords

adversarial training adversarial examples generative adversarial network

Cite

@article{arxiv.2104.13302,
  title  = {Adaptive Adversarial Training for Meta Reinforcement Learning},
  author = {Shiqi Chen and Zhengyu Chen and Donglin Wang},
  journal= {arXiv preprint arXiv:2104.13302},
  year   = {2021}
}

Related papers

View all related →