Noise and Dynamical Synapses as Optimization Tools for Spiking Neural Networks

Standard ANNs lack flexibility when handling corrupted input due to their fixed structure. In this paper, a spiking neural network utilizes biological temporal coding features in the form of noise-induced stochastic resonance and dynamical synapses to increase the model's performance when its parameters are not optimized for a given input. Using the analog XOR task as a simplified convolutional neural network model, this paper demonstrates two key results: (1) SNNs solve the problem that is linearly inseparable in ANN with fewer neurons, and (2) in leaky SNNs, the addition of noise and dynamical synapses compensate for non-optimal parameters, achieving near-optimal results for weaker inputs.