Simultaneous Feature and Model Selection for Continuous Hidden Markov Models.

Hao Zhu,Zhongshi He,Henry Leung
DOI: https://doi.org/10.1109/lsp.2012.2190280
2012-01-01
IEEE Signal Processing Letters
Abstract:In this letter, we propose a novel approach of simultaneous feature and model selection for continuous hidden Markov model (CHMM). In our method, a set of real valued quantities, defined as feature saliencies, are proposed for feature selection. A variational Bayesian (VB) framework is applied to infer the feature saliencies, the number of hidden states, and the parameters of the CHMM simultaneously. Experiments based on synthetic and real data demonstrate the effectiveness of the proposed method.
What problem does this paper attempt to address?