Deep Learning Algorithm Based Support Vector Machines

This paper proposes a new deep learning model which replaces the softmax activation function with support vector machines. To evaluate the performance of the model, we have completed a total of four sets of codes, including the traditional svm classification model, the traditional cnn model, the model of svm behind the fully connected layer, and the model of svm instead of softmax. In order to compare the accuracy of these four groups of models, we trained and tested three data sets, namely the mnist data set, the CIFAR-10 data set and the compass x-ray data set.