基于深度学习下的卷积神经网络参数学习  

Parameter learning of convolution neural networks based on deep learning

作　　者：吕响 张书玉 宋英楠 王长忠[1] 李雯琦 牛嘉瑞 LV Xiang;ZHANG Shuyu;SONG Yingnan;WANG Changzhong;LI Wenqi;NIU Jiarui(College of Mathematical Sciences,Bohai University,Jinzhou 121013,China)

机构地区：[1]渤海大学数学科学学院,辽宁锦州121013

出　　处：《渤海大学学报（自然科学版）》2021年第4期369-375,共7页Journal of Bohai University:Natural Science Edition

基　　金：国家自然科学基金项目(No:61976027);辽宁省科技厅科学技术基金项目(No:XLYC2008002)。

摘　　要：近年来,深度学习(deep learning)在图像目标识别、语音识别和自然语言处理等诸多方面都得到了广泛的应用.目前,深度学习的主要形式是深层神经网络,而深度卷积神经网络(Convolutional neural networks,CNN)则是其中一种经典而广泛应用的结构.随着深度学习方法在诸多领域的不断深入应用,深层卷积神经网络的学习研究在特征学习、目标分类、边框回归等方面表现出的优势已愈发突出.当前关于卷积神经网络的大多参考书目仅仅是说明了原理,却缺少中间参数迭代计算的推导过程,使得数学基础较弱的初学者理解其原理较为困难.本文详细地给出了CNN算法各层的参数的计算公式及其反向传播下的参数的计算公式的推导过程.这有助于加强对网络结构的了解和对网络参数的学习有一个深刻的认识,对以后的学习及建立模型有很大的帮助.In recent years,deep learning has achieved many good results in image target recognition,speech recognition and natural language processing.At present,the main form of deep learning is deep neural network;Convolutional neural networks(CNN)is one of the classical and widely used structures.With the wide application of deep learning methods in many fields,the advantages of deep CNN in feature learning,target classification and border regression have become increasingly prominent.The current bibliographies on CNN only explain the principle,but lack the derivation process of iterative calculation of intermediate parameters,which makes it difficult for beginners with weak mathematical foundation to understand its principle.In view of fact,this paper gives in detail the derivation process of the calculation formula of the parameters of each layer of CNN algorithm and in back propagation algorithms.This helps to strengthen the understanding of network structure and to have a deep insight of network parameters,which is of great help to the future learning and modeling.

关 键 词：深度学习 神经网络 激活函数 反向传播 参数学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]