Learning from Unreliable Trajectories: Adversarially-Robust Federated Q-Learning

We study federated reinforcement learning in which multiple agents interact with a common Markov decision process and communicate through a central server to collaboratively learn the optimal state-action value function.