Monday January 31 4:00 - 5:00 PM Kelley 1001
Saket Joshi Postdoctoral Researcher School of EECS Oregon State University Advances in Decision Theoretic Planning The field of Decision Theoretic Planning (DTP) has been highly successful in providing solutions for problems of planning under uncertainty. This success is largely due to the ability of DTP algorithms to exploit structure and regularities in planning domains. Markov decision process (MDP) have provided the conceptual framework for these algorithms. However, the combinatorial nature of the state spaces often makes the use of classical dynamic programming solutions for MDPs, infeasible. This talk will summarize some of the landmark developments in DTP. Particular focus will be on algorithms that exploit relational structure in planning domains by leveraging First Order Decision Diagrams (FODD), a compact knowledge representation for capturing real valued functions over relational structures. The talk will also briefly touch upon some expressive extensions of FODDs and their application to DTP. Biography Saket Joshi completed his Masters degree in 2003 at the Computer Science department at OSU under Prof. Tom Dietterich. After spending a few months at Microsoft he joined the Computer Science department at Tufts University as a graduate student. He graduated from Tufts in 2010 with a PhD under the guidance of Prof. Roni Khardon. Since October 2010 he has been a post doc with Prof. Prasad Tadepalli at OSU. Saket's primary research interest and main focus has been on the development of agents that can successfully act in complex and uncertain environments. _______________________________________________ Colloquium mailing list [email protected] https://secure.engr.oregonstate.edu/mailman/listinfo/colloquium
