Continuous-time spiking neural networks: general paradigm and event-driven simulation

Abstract

The aim of this research is to develop a simple and effective continuous-time Spiking Neural Network simulator, that takes into account basic biological neuron parameters, in which the latency time is the main effect for the spike generation. A preliminary accurate analysis of the latency time has been developed, applying classical modelling methods to single neurons, by simulations on the most accurate biological model: the Hodgkin-Huxley Model. On the basis of the classical neuron theory, other fundamentals parameters of the systems are defined, such as subthreshold decay, refractory period, inhibitory behaviour, synaptic plasticity, etc. Indeed, spike transmission and latency problems introduce the necessity of using continuous time simulation. Thus, direct use of digital computational methods, seem not completely appropriate. Due to the implicit high-sensitivity of the overall system to close events (conferred by the latency), and the high temporal dynamics of activity, an event-driven simulation method is necessary. In fact, for the proposed neural model, high precision and effectiveness are basically required. A class of fully asynchronous Spiking Neural Networks with a high biological plausibility is definitively proposed, and networks with up to 100.000 neurons can be simulated in a quite short time with a simple MATLAB program. Is also possible to apply plasticity algorithms to emulate interesting global effects, as the Neuronal Group Selection or the jitter-reduction. Moreover, such a parallel processing system could be used for, but not only, engineering problems that involve the use of the classic artificial neural networks (e.g., pattern recognition) . Other applications concern the operation study of biological neural circuits and the exploration of chaotic dynamics in nervous system.