Pairwise decomposition for combinatorial optimization in graphical models
2011
Favier, Aurélie | De Givry, Simon | Legarra , Andres | Schiex, Thomas
We propose a new additive decomposition of probabilitytables that preserves equivalence of the joint distribution while reducing the size of potentials, without extra variables. We formulate the Most Probable Explanation (MPE) problem in belief networks as a Weighted Constraint Satisfaction Problem (WCSP). Our pairwise decomposition allows to replace a cost function with smaller-arity functions.The resulting pairwise decomposed WCSP is then easier to solve using state-of-the-art WCSP techniques. Although testing pairwise decomposition is equivalent to testing pairwise independence in the original belief network, we show how to efficiently test and enforce it, even in the presence of hard constraints. Furthermore, we infer additional information from the resulting nonbinarycost functions by projecting&subtracting them on binary functions. We observed huge improvements by preprocessing with pairwise decomposition andproject&subtract compared to the current state-ofthe-art solvers on two difficult sets of benchmark.
Show more [+] Less [-]Bibliographic information
This bibliographic record has been provided by Fundamental Library of Latvia University of Life Sciences and Technologies
