Autonomy in Learning Sensorimotor Spaces with Dynamic Neural Fields

Abstract

The metaphor of cognition as a dynamical system - which evolves under external forces, is constrained by the internal structure, and is shaped by the agent's behavioral history - is a fruitful source of inspiration guiding experimental and theoretical work. Dynamic Field Theory offers a framework, in which cognitive processes and their development may be modeled quantitatively as dynamics of activation functions defined over behaviorally-relevant parameter spaces. Although learning through memory trace formation is an integral part of the DFT, the behavioral spaces are assumed to be given. However, these spaces may be shaped autonomously while acting in an environment. The autonomy of learning processes is achieved if the behavioral states have intentional structure, which sustains representations to enable learning and ensures their deactivation when appropriate. In this work, I show how intentional structure realized with Dynamic Neural Fields enables autonomous development of a sensorimotor mapping involved in looking behavior.


Back to Table of Contents