《Reducing the Dimensionality of Data with Neural Network》阅读心得

      本论文是2006发表在《SCIENCE》上的一篇关于使用神经网络进行数据降维的文章，作者为G. E. Hinton and R. R. Salakhutdinov。

        论文的思想是使用一个多层的神经网络对数据进行处理，中间层神经元个数少，实现从高维数据到低维数据的‘编码’，然后再通过低维到高维的‘解码’进行数据的重构，目标是减少元数据与重构数据的交叉熵。但是如果权重的初始化不合适时结果会表现不够好例如陷于局部最小，因此在前期先对权重进行初始的训练使得初始值接近最优值，整个项目叫做‘自动编码器’。如图

整个过程分两个阶段：pretraining和unrolling

第一阶段：构造神经元依次减少的受限波兹曼机，前一个rbm输出是后一个rbm的输入，然后根据能量最小得到权重和偏置的迭代公式，其中用到了cd-1方法，在收敛时得到初始训练的权重和偏置。

能量函数：

权重更新：

cd-1可以理解为使用重构的数据再次得到隐藏层偏置

第二阶段：将训练出来的受限波兹曼机依次连接起来进行训练，通过编码解码得到重构数据，目标函数是减少原始数据与重构数据之间的差距，方法使用反向传播调整参数。

注：受限波兹曼机RBM知识：点击打开链接

cd-1是一种采样方法，可参见：随机采样方法

    对于作者的代码的我的注释如下：(注：只是本人的理解可能会有错误，欢迎大家指出来)源代码出处

1.mnistdeepauto。m

% Version 1.000
%
% Code provided by Ruslan Salakhutdinov and Geoff Hinton  
%
% Permission is granted for anyone to copy, use, modify, or distribute this
% program and accompanying programs and documents for any purpose, provided
% this copyright notice is retained and prominently displayed, along with
% a note saying that the original programs are available from our 
% web page. 
% The programs and documents are distributed without any warranty, express or
% implied.  As the programs were written for research purposes only, they have
% not been tested to the degree that would be advisable in any important
% application.  All use of these programs is entirely at the user's own risk.


% This program pretrains a deep autoencoder for MNIST dataset
% You can set the maximum number of epochs for pretraining each layer
% and you can set the architecture of the multilayer net.

clear all
close all

maxepoch=10; %In the Science paper we use maxepoch=50, but it works just fine. 最大迭代数
numhid=1000; numpen=500; numpen2=250; numopen=30;%rbm每层神经元个数1000-500-250-30

fprintf(1,'Converting Raw files into Matlab format \n');
converter; 

fprintf(1,'Pretraining a deep autoencoder. \n');
fprintf(1,'The Science paper used 50 epochs. This uses %3i \n', maxepoch);

makebatches;
[numcases numdims numbatches]=size(batchdata);%每批样本数、维度、批数

fprintf(1,'Pretraining Layer 1 with RBM: %d-%d \n',numdims,numhid);
restart=1;
rbm;
hidrecbiases=hidbiases; 
save mnistvh vishid hidrecbiases visbiases;%保存第1个rbm的权值、隐含层偏置项、可视化层偏置项，为mnistvh.mat

fprintf(1,'\nPretraining Layer 2 with RBM: %d-%d \n',numhid,numpen);
batchdata=batchposhidprobs;
numhid=numpen;
restart=1;
rbm;
hidpen=vishid; penrecbiases=hidbiases; hidgenbiases=visbiases;
save mnisthp hidpen penrecbiases hidgenbiases;

fprintf(1,'\nPretraining Layer 3 with RBM: %d-%d \n',numpen,numpen2);
batchdata=batchposhidprobs;
numhid=numpen2;
restart=1;
rbm;
hidpen2=vishid; penrecbiases2=hidbiases; hidgenbiases2=visbiases;
save mnisthp2 hidpen2 penrecbiases2 hidgenbiases2;

fprintf(1,'\nPretraining Layer 4 with RBM: %d-%d \n',numpen2,numopen);
batchdata=batchposhidprobs;
numhid=numopen; 
restart=1;
rbmhidlinear;
hidtop=vishid; toprecbiases=hidbiases; topgenbiases=visbiases;
save mnistpo hidtop toprecbiases topgenbiases;

backprop; 

converter.m

% Version 1.000
%
% Code provided by Ruslan Salakhutdinov and Geoff Hinton
%
% Permission is granted for anyone to copy, use, modify, or distribute this
% program and accompanying programs and documents for any purpose, provided
% this copyright notice is retained and prominently displayed, along with
% a note saying that the original programs are available from our
% web page.
% The programs and documents are distributed without any warranty, express or
% implied.  As the programs were written for research purposes only, they have
% not been tested to the degree that would be advisable in any important
% application.  All use of these programs is entirely at the user's own risk.
%%将gz格式转为matlab的文件格式
% This program reads raw MNIST files available at 
% http://yann.lecun.com/exdb/mnist/ 
% and converts them to files in matlab format 
% Before using this program you first need to download files:
% train-images-idx3-ubyte.gz train-labels-idx1-ubyte.gz 
% t10k-images-idx3-ubyte.gz t10k-labels-idx1-ubyte.gz
% and gunzip them. You need to allocate some space for this.  

% This program was originally written by Yee Whye Teh 

% Work with test files first 
fprintf(1,'You first need to download files:\n train-images-idx3-ubyte.gz\n train-labels-idx1-ubyte.gz\n t10k-images-idx3-ubyte.gz\n t10k-labels-idx1-ubyte.gz\n from http://yann.lecun.com/exdb/mnist/\n and gunzip them \n'); 
%该文件前四个32位的数字是数据信息  magic number, number of image, number of rows, number of columns
f = fopen('t10k-images-idx3-ubyte','r');
[a,count] = fread(f,4,'int32');
%该文件前两个32位的数字是数据信息  magic number, number of image
g = fopen('t10k-labels-idx1-ubyte','r');
[l,count] = fread(g,2,'int32');

fprintf(1,'Starting to convert Test MNIST images (prints 10 dots) \n'); 
n = 1000;
%Df中存的是.ascii文件代号
Df = cell(1,10);
for d=0:9,
  Df{d+1} = fopen(['test' num2str(d) '.ascii'],'w');
end;
%一次从测试集(1w)中读入1000个图片和标签  rawlabel 1000*1  rawimages 784*1000
for i=1:10,
  fprintf('.');
  rawimages = fread(f,28*28*n,'uchar');
  rawlabels = fread(g,n,'uchar');
  rawimages = reshape(rawimages,28*28,n);
%在对应文档中输入图片的01值(3个整数位)换行
  for j=1:n,
    fprintf(Df{rawlabels(j)+1},'%3d ',rawimages(:,j));%
    fprintf(Df{rawlabels(j)+1},'\n');
  end;
end;

fprintf(1,'\n');
for d=0:9,
  fclose(Df{d+1});
  D = load(['test' num2str(d) '.ascii'],'-ascii');%读取.ascii 中的数据D=内包含样本数*784
  fprintf('%5d Digits of class %d\n',size(D,1),d);
  save(['test' num2str(d) '.mat'],'D','-mat');%转化为.mat文件
end;


% Work with trainig files second  
f = fopen('train-images-idx3-ubyte','r');
[a,count] = fread(f,4,'int32');

g = fopen('train-labels-idx1-ubyte','r');
[l,count] = fread(g,2,'int32');

fprintf(1,'Starting to convert Training MNIST images (prints 60 dots)\n'); 
n = 1000;

Df = cell(1,10);
for d=0:9,
  Df{d+1} = fopen(['digit' num2str(d) '.ascii'],'w');
end;

for i=1:60,
  fprintf('.');
  rawimages = fread(f,28*28*n,'uchar');
  rawlabels = fread(g,n,'uchar');
  rawimages = reshape(rawimages,28*28,n);

  for j=1:n,
    fprintf(Df{rawlabels(j)+1},'%3d ',rawimages(:,j));
    fprintf(Df{rawlabels(j)+1},'\n');
  end;
end;

fprintf(1,'\n');
for d=0:9,
  fclose(Df{d+1});
  D = load(['digit' num2str(d) '.ascii'],'-ascii');
  fprintf('%5d Digits of class %d\n',size(D,1),d);
  save(['digit' num2str(d) '.mat'],'D','-mat');
end;

dos('rm *.ascii');%删除中间文件.ascii

makebatches.m

% Version 1.000
%
% Code provided by Ruslan Salakhutdinov and Geoff Hinton
%
% Permission is granted for anyone to copy, use, modify, or distribute this
% program and accompanying programs and documents for any purpose, provided
% this copyright notice is retained and prominently displayed, along with
% a note saying that the original programs are available from our
% web page.
% The programs and documents are distributed without any warranty, express or
% implied.  As the programs were written for research purposes only, they have
% not been tested to the degree that would be advisable in any important
% application.  All use of these programs is entirely at the user's own risk.
%把数据集及其标签进行打包或分批，方便以后分批进行处理，因为数据太大了，这样可加快学习速率
digitdata=[]; 
targets=[]; 
%训练集中数字0的样本load 将文件中的所有数据加载D上；digitdata大小样本数*784,target大小样本数*10
load digit0; digitdata = [digitdata; D]; targets = [targets; repmat([1 0 0 0 0 0 0 0 0 0], size(D,1), 1)];  
load digit1; digitdata = [digitdata; D]; targets = [targets; repmat([0 1 0 0 0 0 0 0 0 0], size(D,1), 1)];
load digit2; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 1 0 0 0 0 0 0 0], size(D,1), 1)]; 
load digit3; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 0 1 0 0 0 0 0 0], size(D,1), 1)];
load digit4; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 0 0 1 0 0 0 0 0], size(D,1), 1)]; 
load digit5; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 0 0 0 1 0 0 0 0], size(D,1), 1)];
load digit6; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 0 0 0 0 1 0 0 0], size(D,1), 1)];
load digit7; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 0 0 0 0 0 1 0 0], size(D,1), 1)];
load digit8; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 0 0 0 0 0 0 1 0], size(D,1), 1)];
load digit9; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 0 0 0 0 0 0 0 1], size(D,1), 1)];
digitdata = digitdata/255;%累加起来并且进行归一化

totnum=size(digitdata,1);%样本数60000
fprintf(1, 'Size of the training dataset= %5d \n', totnum);

rand('state',0); %so we know the permutation of the training data打乱顺序 randomorder内有60000个不重复的数字
randomorder=randperm(totnum);

numbatches=totnum/100;%批数：600
numdims  =  size(digitdata,2);%维度 784
batchsize = 100;%每批样本数 100
batchdata = zeros(batchsize, numdims, numbatches);%100*784*600
batchtargets = zeros(batchsize, 10, numbatches);%100*10*600

for b=1:numbatches%打乱了进行存储还存在两个数组batchdata，batchtargets中
  batchdata(:,:,b) = digitdata(randomorder(1+(b-1)*batchsize:b*batchsize), :);
  batchtargets(:,:,b) = targets(randomorder(1+(b-1)*batchsize:b*batchsize), :);
end;
clear digitdata targets;

digitdata=[];
targets=[];
load test0; digitdata = [digitdata; D]; targets = [targets; repmat([1 0 0 0 0 0 0 0 0 0], size(D,1), 1)]; 
load test1; digitdata = [digitdata; D]; targets = [targets; repmat([0 1 0 0 0 0 0 0 0 0], size(D,1), 1)]; 
load test2; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 1 0 0 0 0 0 0 0], size(D,1), 1)];
load test3; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 0 1 0 0 0 0 0 0], size(D,1), 1)];
load test4; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 0 0 1 0 0 0 0 0], size(D,1), 1)];
load test5; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 0 0 0 1 0 0 0 0], size(D,1), 1)];
load test6; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 0 0 0 0 1 0 0 0], size(D,1), 1)];
load test7; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 0 0 0 0 0 1 0 0], size(D,1), 1)];
load test8; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 0 0 0 0 0 0 1 0], size(D,1), 1)];
load test9; digitdata = [digitdata; D]; targets = [targets; repmat([0 0 0 0 0 0 0 0 0 1], size(D,1), 1)];
digitdata = digitdata/255;

totnum=size(digitdata,1);
fprintf(1, 'Size of the test dataset= %5d \n', totnum);

rand('state',0); %so we know the permutation of the training data
randomorder=randperm(totnum);

numbatches=totnum/100;
numdims  =  size(digitdata,2);
batchsize = 100;
testbatchdata = zeros(batchsize, numdims, numbatches);
testbatchtargets = zeros(batchsize, 10, numbatches);

for b=1:numbatches
  testbatchdata(:,:,b) = digitdata(randomorder(1+(b-1)*batchsize:b*batchsize), :);
  testbatchtargets(:,:,b) = targets(randomorder(1+(b-1)*batchsize:b*batchsize), :);
end;
clear digitdata targets;


%%% Reset random seeds 
rand('state',sum(100*clock)); 
randn('state',sum(100*clock)); 

rbm.m

% Version 1.000 
%
% Code provided by Geoff Hinton and Ruslan Salakhutdinov 
%
% Permission is granted for anyone to copy, use, modify, or distribute this
% program and accompanying programs and documents for any purpose, provided
% this copyright notice is retained and prominently displayed, along with
% a note saying that the original programs are available from our
% web page.
% The programs and documents are distributed without any warranty, express or
% implied.  As the programs were written for research purposes only, they have
% not been tested to the degree that would be advisable in any important
% application.  All use of these programs is entirely at the user's own risk.

% This program trains Restricted Boltzmann Machine in which
% visible, binary, stochastic pixels are connected to
% hidden, binary, stochastic feature detectors using symmetrically
% weighted connections. Learning is done with 1-step Contrastive Divergence.   
% The program assumes that the following variables are set externally:
% maxepoch  -- maximum number of epochs
% numhid    -- number of hidden units 
% batchdata -- the data that is divided into batches (numcases numdims numbatches)
% restart   -- set to 1 if learning starts from beginning 
%作用：训练RBM，利用1步CD算法 直接调用权值迭代公式不使用反向传播
epsilonw      = 0.1;   % Learning rate for weights 权重学习率 
epsilonvb     = 0.1;   % Learning rate for biases of visible units 可视层偏置学习率
epsilonhb     = 0.1;   % Learning rate for biases of hidden units 隐藏层偏置学习率
weightcost  = 0.0002;   %权衰减，用于防止出现过拟合
initialmomentum  = 0.5;  %动量项学习率，用于克服收敛速度和算法的不稳定性之间的矛盾
finalmomentum    = 0.9;

[numcases numdims numbatches]=size(batchdata);

if restart ==1,%是否为重新开始即从头训练
  restart=0;
  epoch=1;

% Initializing symmetric weights and biases. 初始化权重和两层偏置
  vishid     = 0.1*randn(numdims, numhid);
  hidbiases  = zeros(1,numhid);
  visbiases  = zeros(1,numdims);

  poshidprobs = zeros(numcases,numhid);  %样本数*隐藏层NN数，隐藏层输出p(h1|v0)对应每个样本有一个输出
  neghidprobs = zeros(numcases,numhid);%重构数据驱动的隐藏层
  posprods    = zeros(numdims,numhid);% 表示p(h1|v0)*v0，用于更新Wij即<vihj>data
  negprods    = zeros(numdims,numhid);%<vihj>recon
  vishidinc  = zeros(numdims,numhid);% 权值更新的增量
  hidbiasinc = zeros(1,numhid);% 隐含层偏置项更新的增量
  visbiasinc = zeros(1,numdims);% 可视化层偏置项更新的增量
  batchposhidprobs=zeros(numcases,numhid,numbatches);% 整个数据隐含层的输出  每批样本数*隐含层维度*批数
end

for epoch = epoch:maxepoch,%每个迭代周期
 fprintf(1,'epoch %d\r',epoch); 
 errsum=0;%误差
 for batch = 1:numbatches,  %每一批样本
 fprintf(1,'epoch %d batch %d\r',epoch,batch); 

%%%%%%%%% START POSITIVE PHASE 正向部分%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
  data = batchdata(:,:,batch);%data里是100个图片数据
  poshidprobs = 1./(1 + exp(-data*vishid - repmat(hidbiases,numcases,1)));    %隐藏层输出p(h|v0)
  batchposhidprobs(:,:,batch)=poshidprobs;%将输出存入一个三位数组
  posprods    = data' * poshidprobs;%p(h|v0)*v0 更新权重时会使用到
  poshidact   = sum(poshidprobs);%隐藏层中神经元概率和，在更新隐藏层偏置时会使用到
  posvisact = sum(data);%可视层中神经元概率和，在更新可视层偏置时会使用到

%%%%%%%%% END OF POSITIVE PHASE  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
  poshidstates = poshidprobs > rand(numcases,numhid);%将隐藏层输出01化表示，大于随机概率的置1，小于随机概率的置0，gibbs抽样,设定状态   

%%%%%%%%% START NEGATIVE PHASE  gibbs采样%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
  negdata = 1./(1 + exp(-poshidstates*vishid' - repmat(visbiases,numcases,1)));%01化表示之后算vt=p(vt|ht-1)重构的数据
  neghidprobs = 1./(1 + exp(-negdata*vishid - repmat(hidbiases,numcases,1)));    %ht=p(h|vt)使用重构数据隐藏层的输出
  negprods  = negdata'*neghidprobs;
  neghidact = sum(neghidprobs);
  negvisact = sum(negdata); 

%%%%%%%%% END OF NEGATIVE PHASE %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
  err= sum(sum( (data-negdata).^2 ));%整批数据的误差
  errsum = err + errsum;

   if epoch>5,  %迭代次数不同调整冲量  
     momentum=finalmomentum;
   else
     momentum=initialmomentum;
   end;

%%%%%%%%% UPDATE WEIGHTS AND BIASES 更新权重和偏置%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 
    vishidinc = momentum*vishidinc + ...
                epsilonw*( (posprods-negprods)/numcases - weightcost*vishid);%权重的增量
    visbiasinc = momentum*visbiasinc + (epsilonvb/numcases)*(posvisact-negvisact);%可视层增量
    hidbiasinc = momentum*hidbiasinc + (epsilonhb/numcases)*(poshidact-neghidact);%隐含层增量

    vishid = vishid + vishidinc;
    visbiases = visbiases + visbiasinc;
    hidbiases = hidbiases + hidbiasinc;

%%%%%%%%%%%%%%%% END OF UPDATES %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 

  end
  fprintf(1, 'epoch %4i error %6.1f  \n', epoch, errsum); 
end;

backprop.m

% Version 1.000
%
% Code provided by Ruslan Salakhutdinov and Geoff Hinton
%
% Permission is granted for anyone to copy, use, modify, or distribute this
% program and accompanying programs and documents for any purpose, provided
% this copyright notice is retained and prominently displayed, along with
% a note saying that the original programs are available from our
% web page.
% The programs and documents are distributed without any warranty, express or
% implied.  As the programs were written for research purposes only, they have
% not been tested to the degree that would be advisable in any important
% application.  All use of these programs is entirely at the user's own risk.

% This program fine-tunes an autoencoder with backpropagation.
% Weights of the autoencoder are going to be saved in mnist_weights.mat
% and trainig and test reconstruction errors in mnist_error.mat
% You can also set maxepoch, default value is 200 as in our paper.  
%四个RBM连接起来进行，使用训练数据进行参数的微调整
maxepoch=200;
fprintf(1,'\nFine-tuning deep autoencoder by minimizing cross entropy error. \n');
fprintf(1,'60 batches of 1000 cases each. \n');

load mnistvh%第1个rbm的权值、隐含层偏置项、可视化层偏置项1000
load mnisthp%第二个 500
load mnisthp2%第三个 250
load mnistpo %第四个 30

makebatches;
[numcases numdims numbatches]=size(batchdata);
N=numcases; 

%%%% PREINITIALIZE WEIGHTS OF THE AUTOENCODER初始化权重 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
w1=[vishid; hidrecbiases];%input和第一个rbm 的权值的偏置
w2=[hidpen; penrecbiases];%第一个rbm和第二个rbm
w3=[hidpen2; penrecbiases2];%第二个rbm和第三个rbm
w4=[hidtop; toprecbiases];%第三个和第四个
w5=[hidtop'; topgenbiases]; %第四个和第三个
w6=[hidpen2'; hidgenbiases2]; %第三个和第二个
w7=[hidpen'; hidgenbiases]; %第二个和第一个
w8=[vishid'; visbiases];%第一个和输出

%%%%%%%%%% END OF PREINITIALIZATIO OF WEIGHTS  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

l1=size(w1,1)-1;%每层节点个数
l2=size(w2,1)-1;
l3=size(w3,1)-1;
l4=size(w4,1)-1;
l5=size(w5,1)-1;
l6=size(w6,1)-1;
l7=size(w7,1)-1;
l8=size(w8,1)-1;
l9=l1; %输入层与输出层节点个数相同
test_err=[];
train_err=[];


for epoch = 1:maxepoch

%%%%%%%%%%%%%%%%%%%% COMPUTE TRAINING RECONSTRUCTION ERROR 计算训练误差%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
err=0; 
[numcases numdims numbatches]=size(batchdata);
N=numcases;
 for batch = 1:numbatches
  data = [batchdata(:,:,batch)];
  data = [data ones(N,1)];%每个样本再加一个维度1 是因为w1里既包含权值又包含偏置
  w1probs = 1./(1 + exp(-data*w1)); w1probs = [w1probs  ones(N,1)];
  w2probs = 1./(1 + exp(-w1probs*w2)); w2probs = [w2probs ones(N,1)];
  w3probs = 1./(1 + exp(-w2probs*w3)); w3probs = [w3probs  ones(N,1)];
  w4probs = w3probs*w4; w4probs = [w4probs  ones(N,1)];
  w5probs = 1./(1 + exp(-w4probs*w5)); w5probs = [w5probs  ones(N,1)];
  w6probs = 1./(1 + exp(-w5probs*w6)); w6probs = [w6probs  ones(N,1)];
  w7probs = 1./(1 + exp(-w6probs*w7)); w7probs = [w7probs  ones(N,1)];
  dataout = 1./(1 + exp(-w7probs*w8));
  err= err +  1/N*sum(sum( (data(:,1:end-1)-dataout).^2 )); %剔除掉最后一维
  end
 train_err(epoch)=err/numbatches;

%%%%%%%%%%%%%% END OF COMPUTING TRAINING RECONSTRUCTION ERROR %%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%%%% DISPLAY FIGURE TOP ROW REAL DATA BOTTOM ROW RECONSTRUCTIONS 显示真实的和重构后的数据%%%%%%%%%%%%%%%%%%%%%%%%%
fprintf(1,'Displaying in figure 1: Top row - real data, Bottom row -- reconstructions \n');
output=[];
 for ii=1:15%每次显示15组图片
  output = [output data(ii,1:end-1)' dataout(ii,:)'];%两列真实数据和重构后的数据
 end
   if epoch==1 
   close all 
   figure('Position',[100,600,1000,200]);
   else 
   figure(1)
   end 
   mnistdisp(output);%画图
   drawnow;

%%%%%%%%%%%%%%%%%%%% COMPUTE TEST RECONSTRUCTION ERROR 计算测试数据%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
[testnumcases testnumdims testnumbatches]=size(testbatchdata);
N=testnumcases;
err=0;
for batch = 1:testnumbatches
  data = [testbatchdata(:,:,batch)];
  data = [data ones(N,1)];
  w1probs = 1./(1 + exp(-data*w1)); w1probs = [w1probs  ones(N,1)];
  w2probs = 1./(1 + exp(-w1probs*w2)); w2probs = [w2probs ones(N,1)];
  w3probs = 1./(1 + exp(-w2probs*w3)); w3probs = [w3probs  ones(N,1)];
  w4probs = w3probs*w4; w4probs = [w4probs  ones(N,1)];
  w5probs = 1./(1 + exp(-w4probs*w5)); w5probs = [w5probs  ones(N,1)];
  w6probs = 1./(1 + exp(-w5probs*w6)); w6probs = [w6probs  ones(N,1)];
  w7probs = 1./(1 + exp(-w6probs*w7)); w7probs = [w7probs  ones(N,1)];
  dataout = 1./(1 + exp(-w7probs*w8));
  err = err +  1/N*sum(sum( (data(:,1:end-1)-dataout).^2 ));
  end
 test_err(epoch)=err/testnumbatches;
 fprintf(1,'Before epoch %d Train squared error: %6.3f Test squared error: %6.3f \t \t \n',epoch,train_err(epoch),test_err(epoch));

%%%%%%%%%%%%%% END OF COMPUTING TEST RECONSTRUCTION ERROR %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%组合数据的batches大小由原来的100*600的mini-batches变为1000*60的larger-batches
 tt=0;
 for batch = 1:numbatches/10 
 fprintf(1,'epoch %d batch %d\r',epoch,batch);

%%%%%%%%%%% COMBINE 10 MINIBATCHES INTO 1 LARGER MINIBATCH %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 tt=tt+1; 
 data=[];
 for kk=1:10
  data=[data 
        batchdata(:,:,(tt-1)*10+kk)]; %将10个100行数据连成一行
 end 

%%%%%%%%%%%%%%% PERFORM CONJUGATE GRADIENT WITH 3 LINESEARCHES 共轭梯度%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
  max_iter=3;%3次线性搜索
  VV = [w1(:)' w2(:)' w3(:)' w4(:)' w5(:)' w6(:)' w7(:)' w8(:)']';%将所有的权值和偏置合并为1列
  Dim = [l1; l2; l3; l4; l5; l6; l7; l8; l9];%所有结点

  [X, fX] = minimize(VV,'CG_MNIST',max_iter,Dim,data);
  %VV是权值偏置 CG_MNIST输出的是代价函数和偏导 结点 数据

  w1 = reshape(X(1:(l1+1)*l2),l1+1,l2);
  xxx = (l1+1)*l2;
  w2 = reshape(X(xxx+1:xxx+(l2+1)*l3),l2+1,l3);
  xxx = xxx+(l2+1)*l3;
  w3 = reshape(X(xxx+1:xxx+(l3+1)*l4),l3+1,l4);
  xxx = xxx+(l3+1)*l4;
  w4 = reshape(X(xxx+1:xxx+(l4+1)*l5),l4+1,l5);
  xxx = xxx+(l4+1)*l5;
  w5 = reshape(X(xxx+1:xxx+(l5+1)*l6),l5+1,l6);
  xxx = xxx+(l5+1)*l6;
  w6 = reshape(X(xxx+1:xxx+(l6+1)*l7),l6+1,l7);
  xxx = xxx+(l6+1)*l7;
  w7 = reshape(X(xxx+1:xxx+(l7+1)*l8),l7+1,l8);
  xxx = xxx+(l7+1)*l8;
  w8 = reshape(X(xxx+1:xxx+(l8+1)*l9),l8+1,l9);

%%%%%%%%%%%%%%% END OF CONJUGATE GRADIENT WITH 3 LINESEARCHES %%%%%%%%%%%%%%%%%%%%%%%%%%%%%

 end

 save mnist_weights w1 w2 w3 w4 w5 w6 w7 w8 
 save mnist_error test_err train_err;

end

CG_MNIST.m

% Version 1.000
%
% Code provided by Ruslan Salakhutdinov and Geoff Hinton
%
% Permission is granted for anyone to copy, use, modify, or distribute this
% program and accompanying programs and documents for any purpose, provided
% this copyright notice is retained and prominently displayed, along with
% a note saying that the original programs are available from our
% web page.
% The programs and documents are distributed without any warranty, express or
% implied.  As the programs were written for research purposes only, they have
% not been tested to the degree that would be advisable in any important
% application.  All use of these programs is entirely at the user's own risk.
%得代价函数和对权值的偏导数
function [f, df] = CG_MNIST(VV,Dim,XX);%权值,结点,输入数据
% f ：代价函数，即交叉熵误差
% df ：代价函数对各权值的偏导数
l1 = Dim(1);%各层节点个数
l2 = Dim(2);
l3 = Dim(3);
l4= Dim(4);
l5= Dim(5);
l6= Dim(6);
l7= Dim(7);
l8= Dim(8);
l9= Dim(9);
N = size(XX,1);

% Do decomversion.权值矩阵化
 w1 = reshape(VV(1:(l1+1)*l2),l1+1,l2);%依次取出每层的权值和偏置
 xxx = (l1+1)*l2;%offset
 w2 = reshape(VV(xxx+1:xxx+(l2+1)*l3),l2+1,l3);
 xxx = xxx+(l2+1)*l3;
 w3 = reshape(VV(xxx+1:xxx+(l3+1)*l4),l3+1,l4);
 xxx = xxx+(l3+1)*l4;
 w4 = reshape(VV(xxx+1:xxx+(l4+1)*l5),l4+1,l5);
 xxx = xxx+(l4+1)*l5;
 w5 = reshape(VV(xxx+1:xxx+(l5+1)*l6),l5+1,l6);
 xxx = xxx+(l5+1)*l6;
 w6 = reshape(VV(xxx+1:xxx+(l6+1)*l7),l6+1,l7);
 xxx = xxx+(l6+1)*l7;
 w7 = reshape(VV(xxx+1:xxx+(l7+1)*l8),l7+1,l8);
 xxx = xxx+(l7+1)*l8;
 w8 = reshape(VV(xxx+1:xxx+(l8+1)*l9),l8+1,l9);


  XX = [XX ones(N,1)];
  w1probs = 1./(1 + exp(-XX*w1)); w1probs = [w1probs  ones(N,1)];
  w2probs = 1./(1 + exp(-w1probs*w2)); w2probs = [w2probs ones(N,1)];
  w3probs = 1./(1 + exp(-w2probs*w3)); w3probs = [w3probs  ones(N,1)];
  w4probs = w3probs*w4; w4probs = [w4probs  ones(N,1)];
  w5probs = 1./(1 + exp(-w4probs*w5)); w5probs = [w5probs  ones(N,1)];
  w6probs = 1./(1 + exp(-w5probs*w6)); w6probs = [w6probs  ones(N,1)];
  w7probs = 1./(1 + exp(-w6probs*w7)); w7probs = [w7probs  ones(N,1)];
  XXout = 1./(1 + exp(-w7probs*w8));%输出的概率密度

f = -1/N*sum(sum( XX(:,1:end-1).*log(XXout) + (1-XX(:,1:end-1)).*log(1-XXout)));%代价函数交叉熵
IO = 1/N*(XXout-XX(:,1:end-1));%误差
Ix8=IO; 
dw8 =  w7probs'*Ix8;%向后推导输出层偏导

Ix7 = (Ix8*w8').*w7probs.*(1-w7probs); %第七层残差
Ix7 = Ix7(:,1:end-1);
dw7 =  w6probs'*Ix7;%第七层偏导

Ix6 = (Ix7*w7').*w6probs.*(1-w6probs); 
Ix6 = Ix6(:,1:end-1);
dw6 =  w5probs'*Ix6;

Ix5 = (Ix6*w6').*w5probs.*(1-w5probs); 
Ix5 = Ix5(:,1:end-1);
dw5 =  w4probs'*Ix5;

Ix4 = (Ix5*w5');
Ix4 = Ix4(:,1:end-1);
dw4 =  w3probs'*Ix4;

Ix3 = (Ix4*w4').*w3probs.*(1-w3probs); 
Ix3 = Ix3(:,1:end-1);
dw3 =  w2probs'*Ix3;

Ix2 = (Ix3*w3').*w2probs.*(1-w2probs); 
Ix2 = Ix2(:,1:end-1);
dw2 =  w1probs'*Ix2;

Ix1 = (Ix2*w2').*w1probs.*(1-w1probs); 
Ix1 = Ix1(:,1:end-1);
dw1 =  XX'*Ix1;

df = [dw1(:)' dw2(:)' dw3(:)' dw4(:)' dw5(:)' dw6(:)'  dw7(:)'  dw8(:)'  ]'; 

minimize.m这个文件是用来实现共轭梯度下降进行优化的，但是我没太读懂，如果小伙伴懂的话欢迎交流！