A simulation tool for modeling the influence of anatomy on information flow using discrete integrate and fire neurons

Abstract

There are theories on brain functionality that can only be tested in very large models. In this work, a simulation model appropriate for working with large number of neurons was developed, and Information Theory measuring tools were designed to monitor the flow of information in such large networks. The model’s simulator can handle up to one million neurons in its current implementation by using a discretized version of the Lapicque integrate and fire neuron instead of interacting differential equations. A modular structure facilitates the setting of parameters of the neurons, networks, time and most importantly, architectural changes.

Applications of this research are demonstrated by testing architectures in terms of mutual information. We present some preliminary architectural results showing that adding a virtual analogue to white matter called “jumps” to a simple representation of cortex results in: (1) an increase in the rate of mutual information flow, corresponding to the “bias” or “priming” hypothesis; thereby giving a possible explanation of the high speed response to stimuli in complex networks. (2) An increase in the stability of response of the network; i.e. a system with “jumps” is a more reliable machine. This also has an effect on the potential speed of response.