文档介绍：深度学习算法的研究与实现
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
ABSTRACT
Deep learning has developed rapidly in recent years, and has made breakthrough progress. With the help of the Deep Learning algorithm, human found a way how to deal with the abstract concept in our common problem. This paper firstly introduces the background of the Deep Learning study, and then summarizes the research process and the development of the Deep Learning. Next, this paper introduces the basic principles of deep learning, and on the basis of that, the application of the deep learning was also introduced in some excellent area.
Next, this paper uses the Caffe deep learning platform developed by Berkeley University to carry out the recognition of handwritten numeral characters. The data is from the MNIST database, through the process of generating LMDB format data set, writing deep learning model file, writing solver file, finally by calling the Caffe executable file to training and testing of the deep learning network, we get the final experimental results.
After 10 thousand iterations, the experimental results show that the deep learning network composed of convolutional neural networks can reach more than 99% of the right rate, showing the strong ability of deep learning.
Keywords: Deep Learning, Caffe, Handwritten numeral character recognition
目录
1 摘要 1
2 ABSTRACT 2
3 背景介绍 1
4 研究过程与发展现状 2
研究过程 2
浅层学习和深度学习 2
深度学习与人工神经网络 3
发展现状 5
初始化方法、网络层数和激活函数的选择 5
模型结构 6
学习算法 6
5 深度学习基本原理 8
深度学习的基本思想 8
深度学习训练过程 8
自动编码机 10
卷积神经网络 12
卷积神经网络的结构 13
参数减少与权值共享 14