SAT Based Local Improvement (Research Project)
Funding organization: Austrian Science Fund (FWF)
Project number: (FWF P 32441)
Stefan Szeider (PI)
Structural decomposition is one of the most successful approaches to the solution of hard computational problems, such as for probabilistic reasoning and computational medical diagnosis. Finding a suitable structural decomposition is itself a computationally hard problem. In this project we propose to study a new approach to finding structural decompositions. The idea is to use an exact method (in particular one that is based on satisfiability-solvers) to locally improve a heuristically obtained decomposition. Based on this new idea we will develop new algorithms for the decomposition of graphs and hypergraphs as well as for structural Bayesian Network learning. Our new approach bears the potential of achieving better decompositions for instances that are too large to be handled by exact approaches, and it also provides novel applications for satisfiability solver technology.