Invited speaker

Michael Kearns
Computer and Information Science
Institute for Research in Cognitive Science
University of Pennsylvania


Reinforcement Learning in Dialogue Systems


I will describe our application of reinforcement learning and related
methods in two rather different dialogue systems. The first is a
spoken dialogue system in which learning was applied directly to
low-level parameters of the dialogue manager. The second was a
text-based software agent that provided "social statistics" in a
well-known chat environment. Successes and difficulties of the
approach will be highlighted.