Extensions of the Minimum Discrimination Information Approach for Hidden Markov Modeling
- Dembo A.
- Ephraim Y.
- Rabiner L.
In this paper we extend the minimum discrimination information (MDI) approach for hidden Markov modeling of information sources proposed in , and present an efficient algorithm for its implementation. Furthermore, we provide exact conditions under which the segmental k -means approach, which is an approximation to the maximum likelihood (ML) modeling approach, and the MDI modeling approach, are equivalent in the sense that they lead to the same sequence of models when applied to the same initial model.