卷积神经网络的参数优化和函数选择

Parameter Optimization and Function Selection of Convolutional Neural Networks

卷积神经网络在各大领域都取得了不错的研究成果,并已成熟地应用于许多工业项目,俨然成为了未来人工智能的一种发展趋势.文章首先回顾了卷积神经网络的基本结构和原理,然后主要分析了内部的初始化参数、激活函数与损失函数的选择以及超参数设置对模型训练时间及准确率带来的影响,并在测试了所有的组合方式后挖掘出其中最优的模型.在TensorFlow平台上以MNIST数据集进行验证,其结果表明该模型在不同的需求下都取得了不错的训练结果.

Convolutional neural networks have achieved good research results in various fiel ds,and have been applied to many industrial projects,which has become a development trend of artificial intelligence in the future.This paper first reviews the basic structure and principle of convolutional neural networks,and then analyzes the internal initialization parameters,the selection of activation and loss functions,and the impact of hyperparameter settings on model training time and accuracy.After all combinations,the best model is mined.The MNIST data set was verified on the TensorFlow platform.The results show that the model has achieved good training results under different requirements.