统计与数据科学系系列学术报告之四百五十五期
时 间：2024年11月5日（星期二）14:00-15:00
地 点：史带楼204室
主持人：复旦大学 管理学院 统计与数据科学系 夏寅 教授
报告人：Prof. Cun-Hui Zhang Rutgers University
题 目：Sharp Non-Asymptotic Regret Bounds in Multi-Armed Bandits
摘 要：We present a new approach to the analysis of upper confidence bound (UCB) indices in multi-armed bandits, based on renewal theory, which yields non-asymptotic regret bounds for Gaussian rewards and rewards bounded in [0, 1]. Our analysis leads to regret bounds with sharp leading constant of the main order terms, and only requires finite second moments for inferior arms to achieve logarithmic regret bounds. Our theory also implies that a suitable choice of the exploration function can lead to negative or zero lower order terms under a fixed horizon. Additionally, we provide a non-asymptotic bound for the square-root boundary crossing probability of Brownian motion, which is of independent interest.
个人简介：Cun-Hui Zhang, Distinguished Professor of Statistics at Rutgers University, is a Fellow of the Institute of Mathematical Statistics and a Fellow of American Statistical Association. His research interests include high-dimensional data, machine learning, empirical Bayes, time series, nonparametric methods, multivariate analysis, survival data and biostatistics, functional MRI, closed loop diabetes control, and network to-mography.
统计与数据科学系
2024-10-30
-
活动讲座
-
新闻动态
-
微信头条
-
招生咨询
-
媒体视角
-
瞰见云课堂
