% HebbPatAssoc.m 
% this script trains two-layered binary networks
% to associate patterns using the Hebb rule

[nPat,nIn]=size(InPat); % find number of patterns and number of inputs
[nPat,nOut]=size(DesOut); % find numbers of patterns and outputs
V=zeros(nOut,nIn); % initialize connectivity matrix V to all zeros
for i=1:nOut, % for each output unit
    for j=1:nIn, % for each input unit
        for l=1:nPat, % for each pattern (letter l indexes patterns)
            x=InPat(l,j); % set input x to input pattern
            y=DesOut(l,i); % set output y to desired output
            deltaV=y*x; % compute Hebb weight update
            V(i,j)=V(i,j)+deltaV; % apply weight update
        end % end pattern loop
    end % end input unit loop
end % end output unit loop

Out=zeros(nPat,nOut); % define array to hold output responses
for l=1:nPat, % for each pattern (letter l indexes patterns)
    x=InPat(l,:)'; % set input x to input pattern l
    q=V*x; % find the weighted input sum q 
    y=q>0; % threshold the weighted input sum to find y
    Out(l,:)=y'; % set row l of Out to output y
end % end pattern loop

V % show the trained weight matrix
Out % show the trained output responses


% tighter alternatives to above
% VHb=(DesOut)'*(InPat); % compute weight matrix V in one step
% OutHb=(VHb*InPat')'>0; % compute thresholded output responses in one step
% VPr=(2*DesOut-1)'*(InPat); % compute weight matrix V in one step
% OutPr=(VPr*InPat')'>0; % compute thresholded output responses in one step
% VPo=(DesOut)'*(2*InPat-1); % compute weight matrix V in one step
% OutPo=(VPo*InPat')'>0; % compute thresholded output responses in one step
% VCo=(2*DesOut-1)'*(2*InPat-1) % compute weight matrix V in one step
% OutCo=(VCo*InPat')'>0 % compute thresholded output responses in one step