Request Confirmation Networks: A Cortically Inspired Approach to Neuro-Symbolic Script Execution

Abstract

This thesis examines Request Confirmation Networks (ReCoNs), hierarchical spreading activation networks with constrained top-down/bottom-up recurrency that are proposed as a possible model for cortical activity during execution of neuro-symbolic sensorimotor scripts. ReCoNs are evaluated in the context of the Function Approximator, a showcase implementation that calculates a function value from a handwritten image of the function. Background is provided on biological and artificial neural networks, with emphasis on other biomimetic approaches to machine learning.