I haven't tested MGE with GV on the whole set, but primary training log shows the convergence is slower using GV.
However, I don't know if it's relevent to the statistics of my training data. I mean if the variance of the training data is large, will be perform in an opposite way?
Should we combine MGE training (using HMgeTool) and GV parameter generation (option in HMGens)? I'm wondering if GV will distort the effect of MGE training or not.
Thanks for your help and have a nice day!
Yuan-Fu Liao (廖元甫)
Department of electronic Engineering,
National Taipei University of Technology
Phone:+886-2-27712171 ext. 2247,Fax:+886-2-27317120,Mobile:+886-919-968592yfliao@xxxxxxxxxxx