Manish Saggar, Tekin Mericli, Sari Andoni, Risto Miikkulainen
A single biological neuron is able to performcomplex computations that are highly nonlinear in nature,adaptive, and superior to the perceptron model. A neuron isessentially a nonlinear dynamical system. Its state depends onthe interactions among its previous states, its intrinsicproperties, and the synaptic input it receives. These factors areincluded in Hodgkin-Huxley (HH) model, which describes theionic mechanisms involved in the generation of an action potential. This paper proposes training of an artificial neuralnetwork to identify and model the physiological properties of abiological neuron, and mimic its input-output mapping. An HHsimulator was implemented to generate the training data. Theproposed model was able to mimic and predict the dynamic behavior of the HH simulator under novel stimulationconditions; hence, it can be used to extract the dynamics (in vivoor in vitro) of a neuron without any prior knowledge of itsphysiology. Such a model can in turn be used as a tool forcontrolling a neuron in order to study its dynamics for further analysis.
In Proceedings of International Joint Conference on Neural Networks, Orlando, Florida, August 2007.

Risto Miikkulainen Faculty risto [at] cs utexas edu
Manish Saggar Ph.D. Alumni saggar [at] stanford edu