From forced to natural competition in a biologically plausible neural network

Abstract

In this paper we propose a new unsupervised neural network whose units exhibit intrinsic plasticity and metaplasticity. We describe three versions of the network: The first version is a two-layered neural network with intrinsic plasticity governing the shifting of the activation function, and the pre-synaptic rule altering synaptic weights. In this first version, competition is forced, so that the most activated neuron is set to one and the others to zero. In the second version, competition is not forced and occurs naturally due to inhibition between second layer's neurons. Competition also occurs naturally in the third version whose architecture resembles the one of the internal granular layer of the koniocortex. All versions of our network categorize input patterns similarly to a conventional competitive neural network. © 2013 Springer-Verlag.