Simulation-based Algorithms for Multi-stage Decision Systems: An Empirical Process Theory Approach

DANIEL J. EPSTEIN DEPARTMENT OF INDUSTRIAL AND SYSTEMS ENGINEERING SEMINAR"Simulation-based Algorithms for Multi-stage Decision Systems: An Empirical Process Theory Approach"Dr. Rahul JainAssistant Professor, USC Electrical Engineering-SystemsABSTRACT: I will start by discussing the state of art in Statistical Learning. We will then focus on a particular area called reinforcement learning. Reinforcement learning models are relevant for many multi-stage decision systems and multi-agent systems including autonomous robotic systems, communication networks, medical decision support systems, distributed databases and information retrieval systems, etc. Such systems are typically modeled as a Markov decision process (MDP).We will first look at some classical results related to MDPs. We will then see the computational challenges in such methods. I will then introduce a simulation-based framework for their design and analysis that yields computationally tractable algorithms. The framework depends on new extensions to the empirical process theory to MDPs that I developed. The theory depends on ideas introduced by Kolmogorov and Tihomirov.THURSDAY, OCTOBER 2, 2008, 11:00 AM – 12:00 PM, ANDRUS GERONTOLOGY BLDG ROOM 309

Location: Ethel Percy Andrus Gerontology Center (GER) - 309

Audiences: Everyone Is Invited

Contact: Georgia Lum