Seminar: How to Solve High-Dimensional Reinforcement Learning Problems while Avoiding the Curse of Dimensionality?

Dr. Amir-Massoud Farahmand (McGill University)

Friday, Feb. 7, 2014, 10:30AM-12PM, EV 3.309

Abstract

In the 21st century, we live in a world where data is abundant. We would like to take advantage of this opportunity to make more accurate and data-driven decisions in many areas of life such as industry, healthcare, business, and government. This opportunity has encouraged many machine learning and data mining researchers to develop tools to benefit from data, especially for challenging high-dimensional problems. Nonetheless, the focus of research so far has mostly been about the task of prediction and many complex decision-making problems, in particular the sequential ones, remain almost untouched.

In this talk, I introduce Reinforcement Learning as a computational framework to model sequential decision-making problems. I then propose some theoretically sound data-driven algorithms to solve high-dimensional reinforcement learning problems that avoid the so-called curse of dimensionality. These algorithms apply some of the most successful principles from the modern machine learning theory to the more general context of reinforcement learning. Finally, I showcase the wide-range of applications of reinforcement learning problems by demonstrating how the proposed algorithms are applied to problems in healthcare (HIV management) and robotics (navigation).

Bio

Amir-massoud Farahmand is a postdoctoral fellow at the School of Computer Science, McGill University. He received his PhD from the University of Alberta in 2011. His research interests are in machine learning, reinforcement learning and sequential decision-making problems, robotics, and optimization. Amir-massoud is the recipient of Natural Sciences and Engineering Research Council of Canada (NSERC) postdoctoral fellowship. His work received the University of Alberta’s Department of Computing Science PhD Outstanding Thesis Award for the period of 2011–2012, and has been published in top machine learning (MLJ, NIPS, ICML) and robotics (IROS and ICRA) venues. He will soon join the Robotics Institute, Carnegie Mellon University to continue his postdoctoral research.