How to constrain Neural Network output?

Hi,

For my research purpose, I need to impose a constrain for the output layer. How to constrain the output values?

x = rand([3,100]);
t = rand([2,100]);

net = feedforwardnet([5 10]);
net = configure(net,x,t);
[net tr]=train(net,x,t);

mod=net(x(:,tr.testInd));

sum(mod)
>>ans =

  Columns 1 through 8

    1.2481    1.3790    1.2986    1.3004    1.2728    1.4082    1.2895    1.3355

  Columns 9 through 15

    1.3199    1.3409    1.2890    1.2890    1.3402    1.2568    1.3171

The above code is just an example.

As you can see, the sum of output values beyond 1, but I'd like to constrain the sum of the output values to be exactly equal to 1.

Could anyone help me?

Answer 1

[Aditya Patil]

If you are trying to get outputs as probabilities, then you can use the softmax layer with a neural network. Feedforward network is a simple network that might not be appropriate for your task.