A Neural Network Training Method Based on Adam Adaptive Learning Rate
Aiming at the learning rate adjustment problem in deep neural network training,this paper introduces regularization term to optimize adaptive moment estimation,to improve the training effect of convolutional neural network.Experiments on CIFAR-10 data set show that compared with traditional adaptive moment estimation,the improved adaptive moment estimation based on regularization mechanism can reduce the training time,improve the test and verification accuracy,and reduce the training loss,showing better generalization ability.
neural networkadaptive moment estimationregularizationlearning rate
NETL
NSTL
万方数据
