Elastic hidden Markov model and its application to sequence classification

摘要

In this paper, we describe a new probabilistic sequence model, elastic hidden Markov model (EHMM). The most popular model used to model sequential patterns is the hidden Markov model (HMM). A major shortcoming of the HMM is the assumption that implies that all observations are only dependent on the state generating them, not on neighboring observations. To cope with this problem, the EHMM represents the correlation between observations by adapting the model parameters to an observed sequence on Bayesian framework. Finally, we use EHMMs to model online digit patterns and show that EHMMs can capture the correlation structure in this data set.