Abstract
This paper proposes three methods to improve the learning algorithm for spiking neural networks (SNNs). The aim is to improve learning performance in SNNs where neurons are allowed to fire multiple times. The performance is analyzed based on the convergence rate, the concussion condition in the training period and the error between actual output and desired output. The exclusive-or (XOR) and Wisconsin breast cancer (WBC) classification tasks are employed to validate the proposed optimized methods. Experimental results demonstrate that compared to original learning algorithm, all three methods have less iterations, higher accuracy, and more stable in the training period.