wendy0402
wendy0402

Reputation: 495

calculate error neural network

Recently I have been implementing neural networks for my research. While trying to design the error of the neural network, I got confused on several things because I found several ways to compute mean square error:

  1. global error= sigma of ((Tik-Yik)^2 ) where i= number of output neuron and k= number of training data RMS= sqrt(global error/i+k)

  2. global error= sigma of ((Tik-Yik)^2 ) where i= number of output neuron and k= number of training data mean error= sigma of((Tik-mean(Tik))^2) where i=number of output neuron and k=number of trainingdata RMS=global error/mean error

I got confused about those two, can somebody explain to me which one is the right one? Or both are true?

Upvotes: 0

Views: 1200

Answers (1)

Oke Uwechue
Oke Uwechue

Reputation: 489

Assuming these are the equations you're referring to :

If so, the RMS1 value looks like a valid RMS expression, while RMS2 is simply a fractional expression, not an RMS expression at all.

Just one thing though: it looks like your RMS1 also contains an error/typo: the denominator of the contents of your sqrt() should be simply "k", not "i+k". I think you should simply divide by the training set size, and not include the neuron count in there.

It would be useful if you also cited your references for these equations.

Upvotes: 1

Related Questions