Research on CNN Model Based on Mill's Method

Convolutional neural networks (CNN) have significant advantages in solving artificial intelligence tasks, especially in the field of image classification. As the model of the complicated and large data set, CNN model training time is too long problem increasingly prominent, at present a lot of research work are committed to build a faster convergence speed, CNN model of single training time is shorter, but the model of training have been set in advance, only to achieve the set back propagation training times will stop. Therefore, an improved convolutional neural network based on Mill is proposed, and the corresponding Mill-loss model is designed, which is evaluated on the ResNet series with expanded convolution and the underlying LeNet model. The experimental results show that in the same environment, the CNN based on Mill-loss stops training within a few iterations after convergence, the model training time is significantly shortened and the accuracy is within the acceptable range.