Weight convergence of SpikeProp and adaptive learning rate

Surges during training process are a major obstacle in training a Spiking Neural Network (SNN) using Spike-Prop algorithm and its derivatives [1]. In this paper, we perform weight convergence analysis to understand the proper step size during SpikeProp learning and hence avoid surges during the training process. Using the results of weight convergence analysis, we propose an optimum adaptive learning rate in each iteration which will yield suitable step size within the bounds of convergence condition. The performance of this adaptive learning rate is compared with existing methods via several simulations. It is observed that the use of adaptive learning rate significantly increases the success rate of SpikeProp algorithm along with significant improvement in speed.