151. Adaptive Adversarial Training for Meta Reinforcement Learning
- Author
-
Donglin Wang, Zhengyu Chen, and Shiqi Chen
- Subjects
FOS: Computer and information sciences ,Computer Science - Machine Learning ,Linear programming ,Artificial neural network ,Computer Science - Artificial Intelligence ,Computer science ,Process (engineering) ,business.industry ,Machine learning ,computer.software_genre ,Machine Learning (cs.LG) ,Task (project management) ,Adversarial system ,Artificial Intelligence (cs.AI) ,Robustness (computer science) ,Reinforcement learning ,Artificial intelligence ,business ,computer ,Generative adversarial network - 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.
- Published
- 2021