Provably Efficient Reinforcement Learning with Linear Function Approximation - Chi Jin

Explore provably efficient reinforcement learning with linear function approximation in this 28-minute lecture from the Workshop on Theory of Deep Learning. Delve into sequential decision making, sample efficiency, and value-based algorithms as Chi Jin, a Member of the School of Mathematics at the Institute for Advanced Study, presents cutting-edge research. Examine exploration techniques, including multi-armed bandits and Upper Confidence Bound (UCB), before moving beyond tabular settings to linear function approximation. Investigate linear MDPs and related work in this comprehensive overview of reinforcement learning theory and applications.

Intro
Sequential Decision Making
Reinforcement Learning
Sample Efficiency
Value-based Algorithms
Exploration
Multi-armed Bandits
Upper Confidence Bound (UCB)
Q-learning with UCB
Beyond Tabular Setting
Linear Function Approximation
A Natural Algorithm
Linear MDP
Related Work