A simple probabilistic spiking neuron model with Hebbian learning rules

Traditional spiking neural networks (SNNs) uses simulated spiking neuron models for computation units. Action potentials (APs or spikes) are generated when the integrated sensory or synaptic inputs to a neuron reach a threshold value. However, spiking generation is not a deterministic process, making current models limited for their potentials and applications. Here we consider the effects of adding probabilistic parameters to the spiking neuron model, which controls the synapses established during spiking generation and transmitting. The Hebbian learning rule is employed for controlling the probabilistic parameters self-adaptation and connection weights associated with the synapses are established using Thorpe's rule during the network learning procedure. The proposed framework combines the essence of stochastic characteristics of the cortical neurons in vivo, the biologically plausibility of Hodgkin-Huxley type neuron dynamics, as well as the computational efficiency of integrate-and-fire (I&F) type neurons. A simple simulation acquired following aforementioned instructions (based on Izhivich's SNN model) exhibits more explicit behavior and robust performance than the original model and deterministic network organizations.