Provable Offline Reinforcement Learning: Neural Function Approximation, Randomization, and Sample Complexity

In this talk, Thanh will share some of his recent results on offline reinforcement learning (RL), an RL paradigm for domains where exploration is prohibitively expensive or even implausible, but a fixed dataset of previous experiences is available a priori. Specifically, he will focus on discussing how deep neural networks (trained by (stochastic) gradient descents) and randomization lead to a computationally efficient algorithm that has a strong theoretical guarantee for generalization across large state spaces under mild assumptions of distributional shifts while obtaining a favorable empirical performance. He will conclude with a discussion on future directions to make RL more data-efficient, deployment-efficient, and robust.