【直播】【QuACT系列报告】阮雨霏：Linear Bandits with Limited Adaptivity and...

Motivated by practical needs such as large-scale learning, we study the impact of adaptivity constraints to linear contextual bandits, a central problem in online learning and decision making. Unlike traditional online learning problem which has full adaptivity at a per-time-step scale, our work focuses on the model in which the learning process is executed in parallelization but still wants to achieve optimal performance. In this talk, I will show that in such batch learning model, only ‍‍ batches are needed to achieve the optimal regret. Along the way in the proof, I will introduce the distributional optimal design, which is a natural extension of the optimal experiment design in statistical learning, and introduce our statistically and computationally efficient learning algorithm for the problem, which may be of independent interest.This is joint work with Jiaqi Yang and Yuan Zhou.