Application of constraint networks filtering techniques to truth maintenance systems

A truth maintenance system (TMS) is a general problem solving facility designed to work in tandem with an inference engine. Justification-based TMSs do not have sufficient expressive power for many applications. Logic-based TMSs overcome this limitation but are untractable. Therefore, D. A. McAllester (1980) and K. Forbus and J. de Kleer (1991), have proposed to encode formulas in clauses and to run the efficient Boolean constraint propagation algorithm (BCP), which loses the completeness of the deductions. The author introduces the model of constraint networks (CNs), underlining that the notion of local completeness, called local-consistency in CNs, has been widely treated in CNs. An encoding of McAllester's TMS in a dynamic CN is proposed to be able to use CN techniques. It is shown that, not only can as many deductions be computed as with BCP for the same time-consumption, but also superior performance can be obtained for the local completeness computed for a slightly longer running-time.<>