Model Architectures On Hidden Markov Models

Hidden Markov Models(HMM) as statistical models were brought forward by Baum et al at the later 1960s.Nowadays the HMMs have been advanced from the Discrete Hidden Markov Models(DHMM) to the widely-used Continuous Hidden Markov Models and the Factor Analysed Hidden Markov Models(FAHMM) described in this paper.In this paper,we study the architecture of Hidden Markov Models based on the studied of FAHMM.In order to probe into the essence,we study the links between HMMs and automatas.We know that the HMMs are equal to the Probabilistic Automata. Here we prove that the HMMs are equal to the Fuzzy Automata.The content of this article is on several different sides.1. With the studies of FAHMM,this paper gives an iterative estimating algorithm comparing to the previous estimating algorithm with some reduction on the complexity.Also,we give out another algorithm based on the ANN.It can use for decoding straight forward.2. Based on the links between the HMM and the Probabilistic Automata,we study the fuzzy automata and establish the transformations among different model structures.Through these transformations,we can apply the classical algorithms of HMMs on the automatas and use the automatas to study HMMs.3. We give an example of HMMs used in the fuzzy decision sequence.This is an innovation on the fuzzy system.The three sides are all focused on the model architectures of Hidden Markov Models.With the first side,we know what we can do about the innovation of HMMs in different application domains.With the second, we further approach to the essence of HMMs. At last,the application on the fuzzy system is thus based on the study of the links.