I have created feed-forward backprop Network in NNTOOL of MATLAB. I have selected Trainlm as training function and Learngdm as adaption fucntion. And TANSIG as transfer function.
While training network, I have to define training parameters like epochs, goal, min-grad, max_fail, mu.
What do you mean by mu, min-grad and max_fail?
What are their significance?
mu is the control parameter for the algorithm used to train the neural network. Choice of mu directly affect the error convergence. In case of LMS algorithm, mu is dependent on the maximum eigen value of input correlation matrix.
Mu is the training gain it must be between 0.8-1, in neural network it approximate the inverse of the Hessian matrix which is very complicated function.
detailed explanation at Matlab website..
http://au.mathworks.com/help/nnet/ref/trainlm.html
What is meaning of gradient in artificial neural network (NNTOOL) MATLAB?
Mu stands for momentum constant or momentum parameter which is included in weight update expression to avoid the problem of local minimum. Sometimes network may get stuck to local minimum and convergence does not occur. Range of mu is between 0 and 1.
Mu is the adaptation parameter that is being used in the Levenberg - Marquardt optimization process when calculating the update of the parameters: dX = -(jj+I*mu) \ je
- Badges
- Science topic