Abstract

The work presented in this paper merges the Bienenstock-Cooper-Munro (BCM) learning rule with the Spike Timing Dependant Plasticity (STDP) rule to develop a training algorithm for a multi layer Spiking Neural Network (SNN), stimulated using spike trains. The BCM rule is utilised to modulate the height of the plasticity window, associated with STDP, as a function of the activity of the postsynaptic neurons, and in doing so introduces a correlation between the activity of the postsynaptic neurons and their associated weights. The induced correlation uses the activity of postsynaptic neurons to stabilise the weight values across a multi-layer network causing convergence during training. The training algorithm also includes both exhibitory and inhibitory facilitating dynamic synapses that create a frequency filtering mechanism allowing the information presented to the network to be routed to different neurons. A variable neuron threshold level simulates the refractory period.