Efficient Algorithms for Robust Markov Decision Processes

Abstract

Robust Markov decision processes (MDPs) seek for optimal policies in view of the worst transition kernel from within an ambiguity set that specifies the knowledge about the unknown true Markov process.  Although robust MDPs have emerged as powerful modeling tools, robust MDPs have typically been considered to be intractable, except for special cases where the ambiguity sets are rectangular in both the states and the actions. In this talk, we develop tractable solution techniques for robust MDPs whose ambiguity sets are only required to be rectangular in the states.

Short Bio

Clint Chin Pang Ho is a Research Fellow at Imperial College Business School. He received a BS in Applied Mathematics from the University of California, Los Angeles (UCLA) in 2011, an MSc in Mathematical Modeling and Scientific Computing from the University of Oxford in 2012, and a PhD in computational optimization from the Imperial College London in 2016. Clint studies optimization algorithms and computational methods for structured problems, as well as their applications in operations management, business analytics and machine learning.

Venue

Large Conference Room, O'Reilly Institute