Spiking neural network learning algorithms: Using learning rates adaptation of gradient and momentum steps

摘要

In this paper we propose two learning algorithms for a spiking neural network which encodes information in the timing of spike trains. These algorithms are based on dynamic self adaptation for adapting the gradient learning rates (DS-η) and dynamic self adaptation for adapting the gradient learning rates and momentum (DS-ηα) algorithms. In our proposed algorithm, the optimum value for η was obtained from a parabolic function of error in both of these two algorithms and optimum value for α was obtained from our proposed adaptive algorithm. We performed a selection of benchmark problems to investigate the efficiency of our proposed algorithm. Compared to previously proposed algorithms such as SpikeProp and DS-ηα our algorithms, mod-DS-η and mod-DS-ηα, are faster than other methods in learning of the spiking neural networks.