FunctionApproximationd - %Outputs layer's gradient

Info iconThis preview shows pages 1–2. Sign up to view the full content.

View Full Document Right Arrow Icon
% % Mini Project1 Part2 % clear all n=10; %number of neurons in the hidden layer V1=rand(n,2); V2=-rand(n,2); V=0.5*(V1+V2); % initializing 1st layer's weights W1=rand(1,n+1); W2=-rand(1,n+1); W=0.5*(W1+W2); %initializing output layer's weights. y=zeros(n,1); % y(11,1)=-1; o=0; E=0; N=20; %number of samples between -pi and pi eta=.1; %learning coefficient. epoch=0; Hii=W for epoch=1:1000 E=0; % epoch=epoch+1; for i=1:8*N+1 x(i)=-pi+.25*pi*(i-1)/N; %inputs to the network d(i)=sin(x(i))*cos(2*x(i)); %Desired outputs y(:,1)=bivariateexp(V*([x(i) -1]')); %1 st layer's outputs are evaluated o(i)=bivariateexp(W*([y' -1]')); %2 nd layer's input is evaluated
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Background image of page 2
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: %Outputs layer's gradient delta_L=.5*((d(i)-o(i))*(1-o(i)^2)); for nrn=1:n+1 % we also add a dump neuron for augmanted weight deltaY(nrn)=(1-y(nrn,1)^2)*W(1,nrn)*delta_L; end % step 3 error evaluation E=E + .5*((d(i)-o(i))^2) ; for k=1:n fy(k,1)=.5*(1-y(k,1)^2); if (k==n) fy(n+1,1)=0; end end delta_y = W(1,2)*delta_o*.5*fy(:,1); %W %fy(:,1) %step 5 adjusting weights of output layer W=W + eta*delta_o*[y' -1]; %delta_y %V %step 6 adjusting weights of hidden layer V=V + (eta*delta_y(1:n,1)*[x(i) -1]); end e end m=1:8*N; plot(d,'r') hold on plot(o) p epoch Ea=E/(8*N+1) E...
View Full Document

This note was uploaded on 07/13/2011 for the course ECE 501 taught by Professor Deniz during the Spring '11 term at Istanbul Universitesi.

Page1 / 2

FunctionApproximationd - %Outputs layer's gradient

This preview shows document pages 1 - 2. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online