ABSTRACT:

Reinforcement learning (RL) consists of statistical procedures that use data to determine near-optimal policies for sequential decision-making. It makes use of Markov decision processes, and has applications in numerous areas (e.g., environmental control, robotics, supply chain management, competitive game-playing, industrial process control). A major challenge is the extraordinarily “data-hungry” nature of many methods.

In this talk, we discuss approaches for taming data-hungry procedures in RL, including adaptive and data-efficient methods for policy evaluation in rich state spaces; instance-dependent guarantees that distinguish between “easy” and “hard” problems; and the role of stability in leading to fast convergence.

Based on joint research with: Yaqi Duan, Michael Jordan, Koulik Khamaru, and Eric Xia.

BIO:

Martin Wainwright is the Cecil H. Green Professor in Electrical Engineering and Computer Science and Mathematics at MIT, and affiliated with the Laboratory for Information and Decision Systems and Statistics and Data Science Center. He joined the MIT faculty in July 2022 after spending 20 years in Statistics and EECS at the University of California at Berkeley.

Martin is broadly interested in statistics, machine learning, information theory and optimization.  His work has been recognized by various awards, among them the COPSS Presidents’ Award from the Joint Statistical Societies, the David Blackwell Award from the Institute of Mathematical Statistics, a Section Lecturer from the International Congress of Mathematicians, and a Sloan Foundation Fellowship. He has co-authored several books, including on graphical models, sparse statistical models, and high-dimensional statistics.

Hosted by Professor Baharan Mirzasoleiman