Efficient Representation of Pareto Optimal Solutions for Multi-criteria Markov Decision Processes

 

Title: Efficient Representation of Pareto Optimal Solutions for Multi-criteria Markov Decision Processes

Mr. Josiah Hanna
Undergraduate Student, Computer Science Department, University of Kentucky

Wednesday, November 28, 4pm-5pm, Davis Marksbury Building Auditorium.
Cookies and water will be served at 3:45.


Abstract

Planning under uncertainty is a central problem in developing intelligent autonomous systems. The traditional representation for these problems is a Markov Decision Process (MDP). The MDP model can be extended to a Multi-criteria MDP (MMDP) for planning under uncertainty while trying to optimize multiple criteria. However, due to the trade-offs involved in multi-criteria problems there may be infinitely many optimal solutions. In this talk I will provide a description of MMDPs and Pareto optimality. I will then describe my recent work in efficiently computing a compact subset of solutions that represents the entire set of optimal solutions for bi-objective MDPs.