An Evaluation of Handwriting Digit Recognition Using Multilayer SAM Spiking Neural Network

This paper describes evaluation results of the hand writing digit recognition MNIST using the on-chip trainable multilayer SAM spiking neural network(SAM-SNN). The SAM spiking neuron model is the model that added one more parameter to the LIF neuron model and has a higher resolution ability than the LIF neuron model for implementing onto digital circuitry such as FPGAs. So far, we have proposed a supervised training algorithm for the SAM-SNN and implemented it into FPGAs. In this research, we evaluated the image recognition performance of the SAM-SNN by using the MNIST dataset. As the result, the SAM-SNN performed 99.29% for 60000 training data and 94.52% for 10000 test data. Moreover, the SAM-SNN that used the data made by CNN of ANN performed 99.72% for training data and 97.54% for test data. Decreasing the training rate \(\eta \) made the training progress faster.