A Dreaming Approach to Perceptual Class Delimitation Within the DREAM Architecture

Abstract

Perceptual classes allow cognitive systems to organize the continuous perceptual space they are usually faced with into discrete categories so that planning and long-term prospection can take place. However, for the correct operation of the system it is quite important that the perceptual classes are precisely delimited in order to prevent classification errors that would lead to planning and prediction errors. In this paper we propose a simple but effective mechanism that makes use of the prospective capabilities of a cognitive architecture during a dreaming phase in order to allow it to generate hypotheses on the delimitation of perceptual classes. These hypotheses are later tested as the system interacts with the world when in an awake state.