Function approximation with deep learning.

Hello everyone.

My question is how to use deep Neural Networks to approximate a function using built-in MATLAB functions?

I am trying to approximate a function with 200 input samples and 150 output samples, so it is a weird complex function.

I've decided to construct a NN with 200 inputs and 150 outputs and 3 hidden layers of 90 neurans.

My code:

load D;
%--------------------------------------------------------------------------
D.selection = [1 20000; 2 15000; 3 15000; 4 15000; 5 15000; 6 15000; 7 15000; 8 15000];
[D] = create_dataset(D);
%--------------------------------------------------------------------------
net = newff(D.dataset', D.target' , [90 90 90]);
net.trainparam.epochs = 10000;
net = train(net, D.dataset', D.target');
%--------------------------------------------------------------------------
save('net.mat' , 'net');

Where D.selection is types of data the function could get (i.e. from type 1 take 20000 sample, from type 2 take 15000 sample...)

My problem is with this relatively large dataset and this NN with [90 90 90] neurans training is impossible (my PC went out cold).

I've done it with PYTHON with Tensorflow and it worked really well but for some reasons i want to test it in MATLAB as well (without importing tensorflow or anything else just with matlab).

Could you please help me with this (how to make my code works)?

Best regards.

[amin sari]

Any comments will be greatlly appreciated. I really have no idea how to implement this?