Classification of handwritten digits based on the Izhikevich neuron model

Classification problems are a common problem. The problem of recognizing handwritten digits has always been a problem in machine learning studies. The aim of the study is to show that the pulsed neural network (SNN) model can be used in classification problems such as the single-layer neural network model. For this, the same data set was given to both models as input and the accuracy values obtained as a result of the experiments were compared. In this study, the MNIST dataset, which contains handwritten images of all numbers from 0 to 9, was used. Binary combinations of numbers in the data set are given as input to the biologically more real SNN model, which consists of Poisson neurons in the input layer and Izhikevich neurons in the output layer, and the model is expected to make a correct prediction at the output. The highest accuracy rate of % 99 and the lowest % 51 were obtained in the SNN model. When the results of the experiments with both models were compared, it was seen that the SNN model could be used in classification problems.