Characterization of dynamics and information processing of integrate-and-fire neuron models

Abstract

The integrate-and-fire (IF) model is the most widely used simple spiking neuron model in neuromorphic computing as well as in artificial neural network algorithms. Here, we characterize the dynamics and information processing of IF models using computer simulations and information-theoretic approaches. Neural dynamics is analysed by means of the time evolution of axonal spikes and the phase-plane portrait, and the coding efficiency of a neuron is estimated by the ratio of mutual information of input and output spikes for the binary hidden neural state. The exponential IF model exhibits higher similarity to the biophysical model in both neural dynamics and coding, compared to other IF type models. Electronic circuit simulations based on the simulation programme with integrated circuit emphasis reveal the nonlinear current-voltage characteristics of the IF neuron models and criticalities in the neural networks. Relevant information-theoretic measures indicate that the computational capabilities of neuromorphic devices largely depend on the neuron models. Such an approach combined with the analysis of neural dynamics provides a useful tool to investigate underlying dynamics of mathematical neuron models; it is applicable to the design and evaluation of neuromorphic models.