The Cortical Microcircuit as a Recurrent Neural Network

Abstract

The mammalian neocortex has distributed excitatory and inhibitory connectivity that, together with the integrative properties of pyramidal cells and their strong synaptic plasticity, make it ideally suited to implement a neural network design. This chapter summarizes results from the author’s research, consistent with the hypothesis that the neocortical microcircuit is a recurrent neural network that builds dynamical attractors. According to this paradigm, the units of function of the cortex would be groups of neurons forming ensembles or assemblies through Hebbian synaptic plasticity. The canonical cortical microcircuit would thus be a general-purpose neural network, fine-tuned by experience to solve any optimization problem.