No BSD License  

Highlights from
autoassociative memory

autoassociative memory

by

 

Develop a Matlab program to demonstrate a neural network autoassociative memory.

assoc_mem_demo.m
%assoc_mem_demo.m
%John Wetters
%4/8/09
%autoassociative memory p7-10 Neural Network Design
delete diary.txt
diary diary.txt

close all
clear all


display('Do you want to caluculate weight matrix W with pseudoinverse?')
reply=input('y=yes n=no :   ','s')
if isempty(reply)
    reply = 'y';
end



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%1) generating the alphabetical target vectors
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

load alpha.mat %matrix of 35x26,7x5 alphabetical characters

[m,n] = size(alpha) %check matrix loaded correctly m=35 n=26

%A=alpha(:,1); %A is the first column of alpha
%A to Z are the columns of alpha

%plot out A
A=alpha(:,1); %A is the first column vector
plot_vector7_5(A)
title('7x5 plot of "A"')

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% 2) calculating the weight matrix W with the pseudoinverse
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

pinv_alpha=pinv(alpha);%take the pseudoinverse of alpha to make orthogonal matrix to reduce errors

[m,n] = size(pinv_alpha)

if (strcmp(reply,'y'))
W=alpha*pinv_alpha;%calculate weight vectors W using alpha*pseudoinverse
PlotRealMatrix(W)
title('weight matrix with psuedoinverse')
else
%%%%%%%%%%%%%%%%%%%%%%%
%5) comparing to the results without using the pseudoinverse
%this is a test to see what happens without pseudoinverse
%%%%%%%%%%%%%%%%%%%%%%%
    W=alpha*alpha';
    PlotRealMatrix(W)
title('weight matrix without psuedoinverse')
end


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%3) testing the autoassociative memory without noise
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

error_save=[]

%plot without noise
for i=1:26
 
a=hardlims(W*alpha(:,i));
t=alpha(:,i);
error=sum(abs(t-a)) %calculate number of errors
error_save=[error error_save];

%  plot_vector7_5(t) %plot target
%  title('target')
%  
%  plot_vector7_5(a) %compare to simulated output
%  title('simulated')

end

figure 
plot(error_save)
title('error to target without noise')
xlabel('aphabetic character number')
ylabel('errors in character')

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%4) testing the autoassociative memory with noise
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
error_save=[]
noise_save=[]

noise_range = .3:.01:1;
for noiselevel = noise_range
%plot noise
P = alpha + randn(35,26)*noiselevel;
for i=1:26
    
    in=P(:,i); %input has noise
%     test=W*t;
%     plot_vector7_5(test)
%     title(['with noise without limiting ' num2str(noiselevel)])
    
     a=hardlims(W*in);
%     plot_vector7_5(a)
    %title(['with noise with limiting' num2str(noiselevel)])
    t=alpha(:,i); 
error=sum(abs(t-a)) %calculate number of errors
error_save=[error error_save];
noise_save=[noiselevel noise_save];
end
end

figure 
plot(noise_save,error_save,'rx')
title('error to target vs noise level')
xlabel('noise level')
ylabel('number of errors in target')



diary
edit diary.txt

Contact us