Abstract
Hidden Markov models (HMMs) are usually learned using the expectation maximisation algorithm which is, unfortunately, subject to local optima. Spectral learning for HMMs provides a unique, optimal solution subject to availability of a sufficient amount of data. However, with access to limited data, there is no means of estimating the accuracy of the solution of a given model. In this paper, a new spectral evaluation method has been proposed which can be used to assess whether the algorithm is converging to a stable solution on a given dataset. The proposed method is designed for real-life datasets where the true model is not available. A number of empirical experiments on synthetic as well as real datasets indicate that our criterion is an accurate proxy to measure quality of models learned using spectral learning.
This is a preview of subscription content, log in via an institution to check access.
References
Baker, K.: Singular Value Decomposition Tutorial. Ohio State University (2005)
Boots, B., Gordon, G.: An online spectral learning algorithm for partially observable nonlinear dynamical systems. In: Proceedings of AAAI (2011)
Caelli, T., McCane, B.: Components analysis of hidden Markov models in computer vision. In: Proceedings of 12th International Conference on Image Analysis and Processing, pp. 510â515, September 2003
Davis, R.I.A., Lovell, B.C.: Comparing and evaluating HMM ensemble training algorithms using train and test and condition number criteria. Pattern Anal. Appl. 6(4), 327â336 (2003)
Even-Dar, E., Kakade, S.M., Mansour, Y.: The value of observation for monitoring dynamic systems. In: IJCAI, pp. 2474â2479 (2007)
Glaude, H., Enderli, C., Pietquin, O.: Spectral learning with proper probabilities for finite state automaton. In: Proceedings of ASRU. IEEE (2015)
Hall, A.R., et al.: Generalized Method of Moments. Oxford University Press, Oxford (2005)
Hansen, L.P.: Large sample properties of generalized method of moments estimators. Econometrica: J. Econometric Soc. 50, 1029â1054 (1982)
Hsu, D., Kakade, S.M., Zhang, T.: A spectral algorithm for learning hidden Markov models. J. Comput. Syst. Sci. 78(5), 1460â1480 (2012)
Jaeger, H.: Observable operator models for discrete stochastic time series. Neural Comput. 12(6), 1371â1398 (2000)
Mattfeld, C.: Implementing spectral methods for hidden Markov models with real-valued emissions. CoRR abs/1404.7472 (2014). http://arxiv.org/abs/1404.7472
Mattila, R.: On identification of hidden Markov models using spectral and non-negative matrix factorization methods. Masterâs thesis, KTH Royal Institute of Technology (2015)
Mattila, R., Rojas, C.R., Wahlberg, B.: Evaluation of Spectral Learning for the Identification of Hidden Markov Models, July 2015. http://arxiv.org/abs/1507.06346
Vanluyten, B., Willems, J.C., Moor, B.D.: Structured nonnegative matrix factorization with applications to hidden Markov realization and clustering. Linear Algebra Appl. 429(7), 1409â1424 (2008)
Zhao, H., Poupart, P.: A sober look at spectral learning. CoRR abs/1406.4631 (2014). http://arxiv.org/abs/1406.4631
Author information
Authors and Affiliations
Corresponding authors
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
Âİ 2016 Springer International Publishing Switzerland
About this paper
Cite this paper
Liza, F.F., GrzeĊ, M. (2016). Estimating the Accuracy of Spectral Learning for HMMs. In: Dichev, C., Agre, G. (eds) Artificial Intelligence: Methodology, Systems, and Applications. AIMSA 2016. Lecture Notes in Computer Science(), vol 9883. Springer, Cham. https://doi.org/10.1007/978-3-319-44748-3_5
Download citation
DOI: https://doi.org/10.1007/978-3-319-44748-3_5
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-44747-6
Online ISBN: 978-3-319-44748-3
eBook Packages: Computer ScienceComputer Science (R0)