GA-Neural Network Hybridization

How GA can be hybridized with Neural network (with reference to Matlab).

[Steve Grikschat]

Can you explain a little more? Do you want to GA to select parameters for your neural network? Do you want to fit a response?

[Abul Fujail]

in='input_train.tra';

p=load(in);

p=transpose(p);

net=newff([.1 .9;.1 .9;.1 .9;.1 .9],[7,1], {'logsig','logsig'},'trainlm');

net=init(net);

tr='target_train.tra';

x=load(tr);

x=transpose(x);

net.trainParam.epochs=600;

net.trainParam.show=10;

net.trainParam.lr=0.3;

net.trainParam.mc=0.6;

net.trainParam.goal=0;

[net,tr]=train(net,p,x);

y=sim(net,p);

Some codes are shown above... i have 4 input vector and 1 target vector... i want to get the optimum weight with GA so that the mean square error between target and neural network predicted result is minimum. Please suggest me how the GA can be added with this neural network code..

[thomas lass]

I need the full codes of GA can be hybridized with Neural network

Answer 1

[Greg Heath]

I don't see how they can be combined to an advantage.

Just write the I/O relationship for the net in terms of input, weights and output: y = f(W,x). Then use the Global Optimization toolox to minimize the mean square error MSE = mean(e(:).^2) where e is the training error, e = (t-y) and t is the training goal.

Hope this helps.

Greg

[Du]

It is smart

[Shipra Kumar]

greg how can u write y as a function. i am having similar difficulty while implementing ga-nn. would be glad if u could help

[Greg Heath]

y = B2+ LW*tansig( B1 + IW *x);