A Shallow Neural Network for Simple Image Recognition

Abstract

Taking the fashion mnist data set proposed by zalando research team in 2017 as the object, this paper constructed a "shallow" neural network, whose test result accuracy reached 0.98, and then took the mnist data set as the test object to display the test results. when constructing the model, the loss function, activation function, gradient optimizer, overfitting and learning rate were explained, among which the gradient optimizer and overfitting had the greatest influence on the overall model accuracy.