Technical Report Number
The chapter presents methods for efficiently representing logic formulas in connectionist networks that perform energy minimization. Algorithms are given for transforming any formula into a network in linear time and space and for learning representations of unknown formulas by observing examples of satisfying truth assignments. The relaxation process that underlies networks of energy minimization reveals an efficient hill climbing algorithm for satisfiability problems. Experimental results indicate that the parallel implementation of the algorithm with give extremely good average-case performance, even for large-scale, hard satisfiability problems (randomly generated).
Pinkas, Gadi, "Representing and Learning Propositional Logic in Symmetric Connectionist Networks" Report Number: WUCS-93-9 (1993). All Computer Science and Engineering Research.