基于Adam自适应学习率的神经网络训练方法  

A Neural Network Training Method Based on Adam Adaptive Learning Rate

作　　者：张天中 ZHANG Tianzhong(Zhengzhou University of Industrial Technology,Zhengzhou Henan 451100,China)

机构地区：[1]郑州工业应用技术学院,河南郑州1451100

出　　处：《信息与电脑》2024年第6期44-46,共3页Information & Computer

摘　　要：文章针对深度神经网络训练中的学习率调整问题,通过引入正则化项来优化Adam算法,以提高卷积神经网络的训练效果。在CIFAR-10数据集上的实验表明,基于正则化机制的Adam改进算法相较于传统的Adam优化算法缩短了训练时间,提高了测试和验证准确率,并降低了训练损失,表现出更好的泛化能力。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.

关 键 词：神经网络 Adam优化算法 正则化 学习率 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]