搜索
深度学习算法的研究与实现I1摘要深度学习近几年发展迅速，取得了突破性的进展。借助于深度学习算法，人类终于找到了如何处理“抽象概念”这个亘古难题的方法。本文首先介绍了深度学习的背景，然后对深度学习的研究过程以及发展现状进行了综述。接下来本文介绍了深度学习的基本原理，并在此基础上介绍了深度学习目前应用比较成功的几个领域的应用情况。在本文的最后，本文利用伯克利大学开发的Caffe深度学习平台进行了手写数字字符的识别的项目。数据来源于MNIST数据库，在经历生成符合Caffe的LMDB格式数据集、编写深度学习模型文件、本文编写网络求解文件后，通过调用Caffe可执行文件对深度网络进行训练及测试，并得到最终的实验结果。在经历了1万次迭代后，实验结果显示卷积神经网络构成的深度学习网络可以达到99%以上的识别，显示出深度学习强大的识别能力。关键字：深度学习，Caffe，手写数字识别II2ABSTRACTDeeplearninghasdevelopedrapidlyinrecentyears,andhasmadebreakthroughprogress.WiththehelpoftheDeepLearningalgorithm,humanfoundawayhowtodealwiththeabstractconceptinourcommonproblem.ThispaperfirstlyintroducesthebackgroundoftheDeepLearningstudy,andthensummarizestheresearchprocessandthedevelopmentoftheDeepLearning.Next,thispaperintroducesthebasicprinciplesofdeeplearning,andonthebasisofthat,theapplicationofthedeeplearningwasalsointroducedinsomeexcellentarea.Next,thispaperusestheCaffedeeplearningplatformdevelopedbyBerkeleyUniversitytocarryouttherecognitionofhandwrittennumeralcharacters.ThedataisfromtheMNISTdatabase,throughtheprocessofgeneratingLMDBformatdataset,writingdeeplearningmodelfile,writingsolverfile,finallybycallingtheCaffeexecutablefiletotrainingandtestingofthedeeplearningnetwork,wegetthefinalexperimentalresults.After10thousanditerations,theexperimentalresultsshowthatthedeeplearningnetworkcomposedofconvolutionalneuralnetworkscanreachmorethan99%oftherightrate,showingthestrongabilityofdeeplearning.Keywords:DeepLearning,Caffe,HandwrittennumeralcharacterrecognitionIII目录1摘要......................................................................................................................12ABSTRACT...........................................................................................................23背景介绍..............................................................................................................14研究过程与发展现状..........................................................................................24.1研究过程...................................................................................................24.1.1浅层学习和深度学习....................................................................24.1.2深度学习与人工神经网络............................................................34.2发展现状...................................................................................................54.2.1初始化方法、网络层数和激活函数的选择................................54.2.2模型结构........................................................................................64.2.3学习算法........................................................................................65深度学习基本原理..............................................................................................85.1深度学习的基本思想...............................................................................85.2深度学习训练过程...................................................................................85.3自动编码机.............................................................................................105.4卷积神经网络.........................................................................................125.4.1卷积神经网络的结构..................................................................135.4.2参数减少与权值共享..................................................................145.4.3卷积神经网络总结......................................................................156应用领域............................................................................................................166.1语音识别.................................................................................................166.2图像识别.................................................................................................17IV6.3自然语言处理.........................................................................................176.4搜索广告CTR预估................................................................................187深度学习实战演练............................................................................................197.1实验坏境的搭建.....................................................................................197.1.1深度学习平台..............................................................................197.1.2安装依赖库..................................................................................197.1.3安装Caffe.....................................................................................197.2实验概述.................................................................................................207.3实验过程.................................................................................................217.3.1生成指定的数据格式文件..........................................................217.3.2编写深度网络模型文件..............................................................227.3.3编写网络求解文件......................................................................237.4实验结果.................................................................................................248总结....................................................................................................................269参考文献.