A biologically inspired neural network composed of dissimilar single neuron models

Abstract

A multilayer neural network has been developed that consists of slabs of single neuron models. Each slab is composed of a single type of neutrons, which differs between the slabs. The network was trained using a biologically inspired, Hebbian-like, learning rule on EMG data and good training/testing classification performance was obtained. It was shown that the biologically inspired network, the novel architecture of which is derived from the functionally distinct hypercolumns of neurons in the brain, can be successfully applied on difficult classification tasks.