Schema-based learning: Biologically inspired principles of dynamic organization

Abstract

We propose a generalized framework Schema-based learning (SBL) for the design of complete and integrated adaptive autonomous agents incorporating general principles of adaptive organization e.g., bootstrap coherence and coherence maximization principles. A schema is an evolutionarily or experience-based constructed recurrent pattern of interaction or expectation (perceptual, motor, reactive, and predictive schemas) with the environment, and coherence is a measure of the congruence between the result of an interaction with the environment and the expectations the agent has for that interaction. SBL attempts to provide a general and formal framework independent of the particularities of implementation, thus allowing the design and analysis of a wide variety of agents. SBL allows the growth of increasingly complex patterns of interaction between the agent and its environment from an initially restricted stock of schemas while allowing for efficient learning by confining statistical estimation to a narrow credit assignment space.