A Combined Representation to Refine the Knowledge Using a Neuro-Symbolic Hybrid System applied in a Problem of Apple Classification

Abstract

In this paper we present the model of a Neuro- Symbolic Hybrid System (NSHS) that allows us to refine the knowledge associated to specific problem, for example, in problem of objects classification, where most of the systems of artificial vision use a numeric approach to solve the problem. In order to do this refinement we use one criterion of the NSHS known as, knowledge representation type. The knowledge representation type used in this paper is called combined representation, which is a combination among a local representation and a distributed representation. The proposed NSHS model allows the integration of the numeric and symbolic knowledge in order to obtain refinement knowledge. In this work, numeric knowledge comes from a vision system and symbolic knowledge comes from a human expert in apple classification. We give a brief description of each phase of the proposed model and analysis of the results obtained for every approach (symbolic, connectionist and hybrid) are made. The obtained results demonstrated that, if a lack of knowledge exists, the NSHS model can be used to refine the knowledge.