How to give the fitness function for GA from ann model already developed?

i am having neural network program with 3 input and one output. i want to optimize the output using genetic algorithm. how to give the fitness function for GA from ann model already developed?

Answer 1

[Greg Heath]

What you are asking makes no sense.

Given input/target pairings, a genetic algorithm can be used to optimize the weights and parameters of a network to minimize a, typically, monotonically increasing function of abserror = abs(target-output) (of course crossentropy is an exception to measuring difference with subtraction).

What in your mind is an optimal output? What do you vary to optimize it?

[Prachi Kumari]

I have a doubt along the same lines.

I have developed the ANN model. Is it possible to generate a function which captures the ANN model and which I can insert in the problem function box during GA optimization?

What I wish to achieve is that now since my ANN can predict values etc based on my input variables (I am assuming that the ANN has developed some sort of a formula which relates the input and output), how do I find the optimum result (minimum output values) among these outputs given by ANN using GA?

My ANN uses three input variables and three output variables. There is no direct function available for the relation between the input and output. (Hence the ANN)

Is there any further information that is required (by the answer contributor)?

I don't want to start the whole training the ANN with GA as it seems very complicated to me.