CV-EM and Ag-EM are model estimation algorithms that have higher
generalization ability than EM.
They can estimate precise models
from a limited amount of training data. The details of the algorithms
are found here.
A simple sample program that trains and evaluates GMM using EM, CV-EM, and Ag-EM.
The source code can be browsed from here.
A pache file to the HTK toolkit and some scripts to train HMM using CV/Ag-EM.
Please let me know if you have a question.
[back to home]