IMI Interdisciplinary Mathematics InstituteCollege of Arts and Sciences

How to evolve a neuron

  • March 20, 2017
  • 1:15 p.m.
  • LeConte 312

Abstract

Artificial life (a-life) is an application of mathematics and computer simulation in which abstract virtual organisms (agents) perform simplified life-like functions. A-life simulations often include an evolutionary component. I will present one of my a-life projects, in which agents must evolve neuron-like behavior. The virtual machine within each agent consists of an abstraction of a biochemical reaction network that is readily encoded in a binary genome. Agents are scored based on their ability to process an incoming stream of spikes and produce output spikes of their own under certain circumstances. Specifically, they must fire when two input spikes arrive at almost the same time, mimicking the ability of natural neurons to detect coincidences and synchrony. Furthermore, they must adapt and fire more readily after a period of high activity, and become more reluctant to fire after a period of low activity, mimicking neural plasticity. It is particularly challenging to specify a scoring method that causes populations to evolve this behavior in a reasonable amount of time. I will explain the scoring method, describe some of the solutions found by the simulation, and describe plans for future applications of these results.

© Interdisciplinary Mathematics Institute | The University of South Carolina Board of Trustees | Webmaster
USC