Research Post
In this work, we study algorithms for learning in infinite-horizon undiscounted Markov decision processes (MDPs) with function approximation. We first show that the regret analysis of the Politex algorithm (a version of regularized policy iteration) can be sharpened from O(T^3/4) to O(√T^3/4) under nearly identical assumptions, and instantiate the bound with linear function approximation. Our result provides the first high-probability O(√T) regret bound for a computationally efficient algorithm in this setting. The exact implementation of Politex with neural network function approximation is inefficient in terms of memory and computation. Since our analysis suggests that we need to approximate the average of the action-value functions of past policies well, we propose a simple efficient implementation where we train a single Q-function on a replay buffer with past data. We show that this often leads to superior performance over other implementation choices, especially in terms of wall-clock time. Our work also provides a novel theoretical justification for using experience replay within policy iteration algorithms.
Feb 15th 2022
Research Post
Read this research paper, co-authored by Amii Fellow and Canada CIFAR AI Chair Adam White: Learning Expected Emphatic Traces for Deep RL
Feb 15th 2022
Research Post
Read this research paper, co-authored by Canada CIFAR AI Chair Kevin Leyton-Brown: The Perils of Learning Before Optimizing
Feb 14th 2022
Research Post
Read this research paper, co-authored by Amii Fellows and Canada CIFAR AI Chairs Osmar Zaïane,and Lili Mou, Non-Autoregressive Translation with Layer-Wise Prediction and Deep Supervision
Looking to build AI capacity? Need a speaker at your event?