Proceedings Abstracts of the Twenty-Fourth International Joint Conference on Artificial Intelligence

Compiling Constraint Networks into Multivalued Decomposable Decision Graphs / 332
Frédéric Koriche, Jean-Marie Lagniez, Pierre Marquis, Samuel Thomas

We present and evaluate a top-down algorithm for compiling finite-domain constraint networks (CNs) into the language MDDG of multivalued decomposable decision graphs. Though it includes Decision-DNNF as a proper subset, MDDG offers the same key tractable queries and transformations as Decision-DNNF, which makes it useful for many applications. Intensive experiments showed that our compiler cn2mddg succeeds in compiling CNs which are out of the reach of standard approaches based on a translation of the input network to CNF, followed by a compilation to Decision-DNNF. Furthermore, the sizes of the resulting compiled representations turn out to be much smaller (sometimes by several orders of magnitude).