I built a neural network using neural network tool box. I want a function from the built network which I can use in genetic algorithm as objective function. Is there any way ? TIA

I built a neural network using neural network tool box. I want a function from the built network which I can use in genetic algorithm as objective function. Is there any way ? TIA I need a function or equation from my neural network which i trained and tested. now I want to use optimization tool like GA using this network. To use GA i need an objective or fitness function, is there any way I can have this function from neural network ?

Answer 1

[Greg Heath]

The analytic form of a default single hidden node MLP is (tansig == tanh)

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

 y = crossentropy(B2 + LW*tansig(B1+IW*x)) % PATTERNNET

Hope this helps.

Greg

[Saad]

can I directly use this as objective function kindly describe more, it will be much helpful. (I not fluent on this topic). TIA

[Saad]

like I have three inputs and one corresponding output, using these three I/O; I trained the network. After the construction of the net i used the net in order to predict response. Now I want to use that particular net which I created as an objective function on GA. please help me out.

[Greg Heath]

I think you are confused.

The GA is used to design the network. For example, the most typical objective function is mse(output-target) which is to be minimized. Target is given and output is analytically expressed in terms of the unknown weights and biases.

There are many examples in both the NEWSGROUP and ANSWERS.

                  NEWSGROUP     ANSWERS

 neural genetic      137           73
 neural ga            73           59

Hope this helps

Greg