Abstract:

Transfer learning provides a powerful tool for incorporating multiple related studies into a target study of interest with successful applications in machine learning and biological research. In this talk, I will first introduce the similarity characterization of related tasks and transfer learning algorithms for high-dimensional linear regression. Its theoretical guarantees and minimax optimality will be demonstrated. Next, I will introduce a transferred Q-learning algorithm, which can integrate source data into a target offline or online reinforcement learning problem. Improvement in policy learning will be demonstrated theoretically and numerically.

Biography:

李赛，中国人民大学统计与大数据研究院助理教授，博士生导师。2018年于罗格斯新泽西州立大学获得统计博士学位，毕业后于宾夕法尼亚大学生物统计系和统计系进行博士后研究，目前的研究方向包括高维数据分析、迁移学习、因果推断的统计方法及理论和在遗传学、流行病学和机器学习中的应用。

Tencent Meeting：https://meeting.tencent.com/dm/GNrPMMiWBJte

Meeting ID：106 470 325