A new limited search approach to learning Abbadingo-style finite state automata.

摘要

This thesis introduces and evaluates a new limited search approach to learning Abbadingo-style DFA. The new approach resulted from an analysis of the effectiveness of the leading merge selection heuristic. With a simple breadth-first traversal of the available first merge choices of the inference process, the new limited search approach obtains more favourable results than those of the current state of the art solution. These results were acquired from evaluating both algorithms on a new 300 problem benchmark data set that lies slightly beyond one of the most difficult solvable benchmark problems in the well-known Abbadingo One Learning Competition.