Donald Bren School of Information and Computer Sciences

This paper presents an approach to translating the data associated with a problem-solving procedure into efficient implementations. The approach involves reduction of problem-domain data strutures to implementations by way of intermediate modelling structures. Formalists are introduced for describing modelling structures -- abstract representations of data characteristics and relationships, independent of any specific implemetation -- and implementation structures -- machine-processable represetations. Based upon these formalisms, algorithms are presented for recognizing known modelling structures, for synthesizing implementations for modelling structures not recognized, and for combining several implementations according to structure membership and variable binding relationships. Design considerations influencing these formalisms and algorithms are discussed.