Hi I am using a genetic algorithm to optimise the intial weights in a feed-forward neural network, which I will later train using a bp with momentum algorithm. The fitness function I am presently using is 1-MSE (1-mean_square_error). MSE is calculated on a small training set. But this fitness function is giving me very little spread. Another fitness function I came across in the literature is something like 1/MSE. Are there any better fitness functions that I can use? (the 1/MSE thing will grow really large as the error decreases) thanks