An Improved Izhikevich Neuron Model With Time Delay and Adaptive Parameters for High-Performance Spiking Neural Networks

The Izhikevich neuron model is a biophysically accurate model that uses a 2-D system of differential equations. Within these equations are dimensionless variables and parameters. These parameters, selected by the researcher, remain fixed. We propose allowing some of these parameters to adapt based on the firing rate of the neuron. Additionally, we propose introducing a small time delay to make the model more biologically realistic. We also propose combining both adjustments to alter the model. Our experimental results show that adaptive parameters substantially reduce computation time while maintaining competitive accuracy. Additionally, the introduction of time delay preserved the accuracy of the original model while slightly reducing computation time. Authors may consider introducing adaptivity to the time delay feature in future work.