NEOCORTEX'S ARCHITECTURE OPTIMIZES COMPUTATION, INFORMATION TRANSFER AND SYNCHRONIZABILITY, AT GIVEN TOTAL CONNECTION LENGTH

Abstract

It is experimental evidence that biological neocortical neurons are arranged in a columnar clustered architecture and coupled according to a bi-power law connection probability function. We propose a computational framework that relies on recurrent connectivity and work out associated computational properties. Taking the columns as the basic computational elements, we investigate a bi-power connection probability function paradigm, in order to scan a wide range of network types, for which we measure the speed of information propagation and synchronizability. Whereas the speed of information propagation increases linearly in the neighbor order n for n-nearest neighbor coupled networks, for close-to-biology bi-power models of the cortex it quickly saturates at high values, expressing the superiority of this network type. Our results reveal that these networks optimize information propagation and synchronizability at a minimal total connection length.