Knowledge Representation by Inner-class Model and its Application for Pattern Recognition

Abstract

The inner-class model, aims flexible recognition like human, is introduced. The feature is defined on the input signal where the dimension is decreased only in the area do not concerned, meanwhile the dimension is not decreased in the concerning area. The system records every feature, referred to as inner-class, is an element of the knowledge. Since the physical relationship between co-occurring features is expressed by the feature, any meaning of the relation between inner-classes is expressed by a sole relation between them. Also, the meaning of transitive relation among inner-classes is expected to be expressed by the sole relation. The knowledge is trained by increasing the strength of the relation if it is used and by decreasing the strength of the relation if it is not used. One of the applications of this model is pattern recognition. A simple one-shot learning is examined to know proposing model works as it is expected.