Rule compilation from constraint-based problem solving

Abstract

A constraint-based system for automating the acquisition of problem-solving knowledge is described. The approach is novel in attempting to compile rules from the observation of constraint-based, relaxation-based problem solving. The system has three main components; a constraint-based problem solver, a rule-compiler and a rule-base problem solver. A relation consistency algorithm is the backbone of the constraint-based problem solver. One advantage of this method is that customized expert systems can be built by manipulating the problems used for learning. Experiments were performed to evaluate a prototype learning system and some extensions.<>