Toward Automated Rational Reconstruction: A Case Study

Abstract

This chapter reviews a sequence of rational reconstructions of a particular design/learning system called CPS/SCALE that characterize that system's development. CPS is a design system for implementing Boolean functions as gate networks, while SCALE is a learning system that adds refinement rules to CPS's knowledge base. Knowledge-based systems can be represented so their development over time can be expressed as a rational reconstruction process CPS is a search-based design program that performs decomposition, implementation, patching, and backtracking. It contains a knowledge base of implementation rules that map Boolean functions onto gate network implementations. CPS performs (recursive) decomposition of the Boolean function design problem and direct implementation of primitive subproblems, accompanied by patching and chronological backtracking. CPS also contains a knowledge base of patching rules that map gate networks into functionally equivalent gate networks. CPS is the problem solver of a simplified, LEX-like architecture whereas SCALE carries out the functions of critic, generalizer, and problem generator. CPS/SCALE began its life as a perturbation of the DONTE circuit design system.