A Model for Empirical Validation in Self-Updating Cognitive Representation

摘要

It is evident that the human cognitive mechanism exhibits a degree representational plasticity. We consider from an empirical perspective the extent to which fully-autonomous updating of an artificial cognitive agent's perceptual framework can be achieved, given that that the ability to objectively validate both representational strategies as well as represented aspects of the environment must be maintained throughout representational transitions. In particular, it is argued that the problem is intractable unless there exist mechanisms for maintaining an empirical distinction between the various possibilities of perceptual characterization of the environment and the inherent uncertainties in environment modeling. Doing so, we shall argue, requires an a priori percept-action coupling to be retained throughout the update process. A corresponding mechanism for open-ended perception-action learning is described within a first-order logical context for embodied cognitive agents. Experimental results within in a simple simulated puzzle domain suggest that active perceptual updating in this manner can significantly accelerate learning of the of the objective world-model.