Parameter Estimation of Hidden Markov Models (HMM) using go with the Winner Algorithms

Abstract

Hidden Markov model (HMM) is a stochastic method which has been used in various application like speech processing, signal processing and character recognition. It has three main problems. Third problem of HMM is the one in which we optimize the model parameters so as to describe how a given observation sequence comes about. The observation sequence is used to adjust the model parameters is called training sequence since it is used to train the HMM. One of the conventional methods that are applied in setting HMM model parameters values is Baum Welch algorithm. So in this paper Go With the Winner (GWW) method is used to train the HMM Parameters. We have already done experiment of same set of data using Baum Welch, Metropolis, Simulated Annealing and Genetic algorithm. The experimental results show that GWW is found to reach maxima in less number of transactions and the value of P(O|λ) is also much higher in comparison to Metropolis, Simulated Annealing and Genetic algorithm.