Knowledge verification of machine-learning procedures based on test structure measurements

Abstract

The authors describe an approach for evaluating and refining the rules, based on test structure measurements, to be entered into the knowledge base of an expert system that characterizes device performance. The objective is to qualify the performance of rules determined by a machine-learning classification application with the best knowledge available from the human experts. The technique combines a machine-learning approach with the traditional heuristic-based development of an expert system. Strengths and weaknesses of the individual techniques are compared.<>