Approximating MAP using Local Search

MAP is the problem of finding a most probable instantiation of a set of variables in a Bayesian network, given (partial) evidence about the complement of that set. Unlike computing priors, posteriors, and MPE (a special case of MAP), the time and space complexity of MAP is not only exponential in the network treewidth, but also in a larger parameter known as the "constrained" treewidth. In practice, this means that computing MAP can be orders of magnitude more expensive than computing priors, posteriors or MPE. For this reason, MAP computations are generally avoided or approximated by practitioners. We have investigated the approximation of MAP using local search. The local search method has a space complexity which is exponential only in the network treewidth, as is the complexity of each step in the search process. Our experimental results show that local search provides a very good approximation of MAP, while requiring a small number of search steps. Practically, this means that the average case complexity of local search is often exponential only in treewidth as opposed to the constrained treewidth, making approximating MAP as efficient as other computations.

[1]  Adnan Darwiche,et al.  Inference in belief networks: A procedural guide , 1996, Int. J. Approx. Reason..

[2]  Adnan Darwiche,et al.  Recursive conditioning , 2001, Artif. Intell..

[3]  David J. Spiegelhalter,et al.  Probabilistic Networks and Expert Systems , 1999, Information Science and Statistics.

[4]  D. C. Wilkins,et al.  Stochastic Greedy Search: Efficiently Computing a Most Probable Explanation in Bayesian Networks , 2000 .

[5]  R. Dechter,et al.  Stochastic Local Search for Bayesian Networks , 1999 .

[6]  Nevin Lianwen Zhang,et al.  Exploiting Causal Independence in Bayesian Network Inference , 1996, J. Artif. Intell. Res..

[7]  Judea Pearl,et al.  Probabilistic reasoning in intelligent systems , 1988 .

[8]  Gregory F. Cooper,et al.  The Computational Complexity of Probabilistic Inference Using Bayesian Belief Networks , 1990, Artif. Intell..

[9]  Michael Luby,et al.  Approximating Probabilistic Inference in Bayesian Belief Networks is NP-Hard , 1993, Artif. Intell..

[10]  Adnan Darwiche,et al.  A differential approach to inference in Bayesian networks , 2000, JACM.

[11]  Rina Dechter,et al.  Bucket elimination: A unifying framework for probabilistic inference , 1996, UAI.

[12]  Rina Dechter,et al.  Stochastic local search for Bayesian network , 1999, AISTATS.

[13]  José A. Gámez,et al.  Partial abductive inference in Bayesian belief networks using a genetic algorithm , 1999, Pattern Recognit. Lett..

[14]  David J. Spiegelhalter,et al.  Local computations with probabilities on graphical structures and their application to expert systems , 1990 .

[15]  Adnan Darwiche,et al.  Any-Space Probabilistic Inference , 2000, UAI.

[16]  Uue Kjjrull.  Triangulation of Graphs { Algorithms Giving Small Total State Space Triangulation of Graphs { Algorithms Giving Small Total State Space , 1990 .