Using inference to solve reinforcement learning and control problems
By Matt Hoffman
A number of techniques have been proposed recently to transform the problems of
reinforcement learning and control into problems of maximum-likelihood and
inference. This reformulation allows us to use a greater set of tools (such as
EM, MCMC, etc.) to solve these problems and should provide additional insight
into the behavior of some of the more classical algorithms. In this talk I will
present an overview of this methodology and detail more specifically some of the
work I've done in this area.