A multi-criteria exemplar model for holistic categorization in autonomous agents

When developing a model of perceptual categorization, a functional and holistic approach is required. Therefore, considering the influence of an agent's current needs, perceptual categorization is modeled as the valuation of a stimulus regarding its potential to satisfy an agent's needs. Based on an exemplar model, an activation-based model is developed that considers two kinds of categorization criteria, namely perceptual similarity and expectation, which represents bottom-up and top-down approaches of perception, respectively. The aggregation of these two criteria then corresponds to the integration of these approaches into a holistic model. A simplified simulation of the categorization model shows that the multi-criteria model leads to a more confident valuation of stimulus objects. In particular, the model is able to eliminate ambiguity in perceptual categorization. Although used equivalently, the simulation shows that the bottom-up similarity criterion is more significant than the top-down expectation-based criterion. In particular, the latter is only significant in case of ambiguity in an object's appearance.