A Parameterized Local Consistency for Redundant Modeling in Weighted CSPs

The weighted constraint satisfaction problem (WCSP) framework is a soft constraint framework which can model many real life optimization or overconstrained problems. While there are many local consistency notions available to speed up WCSP solving, in this paper, we investigate how to effectively combine and channel mutually redundant WCSP models to increase constraint propagation. This successful technique for reducing search space in classical constraint satisfaction has been shown non-trivial when adapted for the WCSP framework. We propose a parameterized local consistency LB(m,Φ), which can be instantiated with any local consistency Φ for single models and applied to a combined model with m sub-models, and also provide a simple algorithm to enforce it. We instantiate LB(2,Φ) with different state-of-the-art local consistencies AC*, FDAC*, and EDAC*, and demonstrate empirically the efficiency of the algorithm using different benchmark problems.