Learning Weighted Prototypes using Genetic Algorithms

Smith and Osherson proposed the prototype view for concept representation and category cls, Hiflcafion. In the prototype view, concepts are represented as prototypes. A prototype is a collection of salient properties of a conceptUnder the prototype view, a~ instance is classified as a memberofa concept ifit is sufficiently similar to the prototype of this concept. Although the prototype view has been extensively re~azched in cognitive sciest~, it has not been widely adopted in machine learning. In this paper, we discuss some preliminary work on a genetic algorithm.q approach to learning weighted prototypes. In this approach, a concept is represented .as one or more weighted prototypes, each of which is a conjunction of weighted attribute values. In this approach, eve~ prototype maintains its own at~bute weights. A genetic algorithm is applied to generate prototypes and their attribute weight& This approach has beca implemented in GABWPL (Genetic Algorithm BA_~_~!_ Weighted Prototype Leeraing)end empirically evaluated on several artificial datasets.