PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Probabilistic inference for solving (PO)MDPs
Marc Toussaint, Stefan Harmeling and Amos Storkey
(2006) Technical Report. University of Edinburgh, Edinburgh, UK.

This is the latest version of this eprint.


The development of probabilistic inference techniques has made considerable progress in recent years, in particular with respect to exploiting the structure (e.g., factored, hierarchical or relational) of discrete and continuous problem domains. We show that these techniques can be used also for solving Markov Decision Processes (MDPs) or partial observable MDPs (POMDPs) when formulated in terms of a structured dynamic Bayesian network (DBN). The approach is based on an equivalence between maximization of the expected future return in the time-unlimited MDP and likelihood maximization in a related mixture of finite-time MDPs. This allows us to use expectation maximization (EM) for computing optimal policies, using arbitrary inference techniques in the E-step. Unlike previous approaches we can show that this actually optimizes the discounted expected future return for arbitrary reward functions and without assuming an ad hoc finite total time. We first develop the approach for standardMDPs and demonstrate it using exact inference on a discrete maze and Gaussian belief state propagation in non-linear stochastic optimal control problems. Then we present an extension for solving POMDPs. We consider an agent model that includes an internal memory variable used for gating reactive behaviors. Using exact inference on the respective DBN, the EM-algorithm solves complex maze problems by learning appropriate internal memory representations.

PDF - Requires Adobe Acrobat Reader or other PDF viewer.
EPrint Type:Monograph (Technical Report)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Computational, Information-Theoretic Learning with Statistics
Learning/Statistics & Optimisation
Theory & Algorithms
ID Code:3924
Deposited By:Amos Storkey
Deposited On:25 February 2008

Available Versions of this Item