Improving Reinforcement Learning with Human Input

 

Matthew E. Taylor headshot Matthew E. Taylor
Assistant Professor
Allred Distinguished Professorship in Artificial Intelligence
School of Electrical Engineering and Computer Science
Washington State University

June 6, 2017
3:00–4:00pm
Science Hall 107
Host: William Yeoh

Abstract

Reinforcement learning (RL) has had many successes, from controlling video games and robots to web server and data center optimization. However, significant amounts of time and/or data can be required to reach acceptable performance. If agents or robots are to be deployed in real-world environments, it is critical that our algorithms take advantage of existing human knowledge. This talk will discuss a selection of recent work that improves RL by leveraging 1) demonstrations and 2) reward feedback from imperfect users, with an emphasis on how interactive machine learning can be extended to best leverage the unique abilities of both computers and humans.

Biography

Matthew E. Taylor received his doctorate from the Department of Computer Sciences at UT-Austin in the summer of 2008, supervised by Peter Stone. Matt then completed a two-year postdoctoral research position at the University of Southern California with Milind Tambe and spent 2.5 years as an assistant professor at Lafayette College in the computer science department. He currently holds the Allred Distinguished Professorship in Artificial Intelligence at Washington State University in the School of Electrical Engineering and Computer Science and is a recipient of the National Science Foundation CAREER award. Research interests include intelligent agents, multi-agent systems, reinforcement learning, transfer learning, and robotics.