Extended Tied-Mixture HMMs for Both Labeled and Unlabeled Time Series Data

An insu ciency of training data often results in a poorly learned classi er. To mitigate this problem, several learning methods using both labeled and unlabeled data have been proposed. In these methods, however, only static data are considered; time series unlabeled data cannot be utilized. In this paper, we rst present an extension of HMMs, named Extended Tied-Mixture HMMs (ETM-HMMs) in which both labeled and unlabeled time series data can be used simultaneously to obtain a better classication accuracy than the case only labeled data are used. The learning algorithm for the ETM-HMMs is also presented. Experiments on synthetic and gesture data demonstrated that unlabeled time series data can help improve the classi cation performance.