A minimum frame error criterion for hidden markov model training

Taemin Cho, Kibeom Kim, Juan P. Bello

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Hidden Markov models (HMM) have been widely studied and applied over decades. The standard supervised learning method for HMM is maximum likelihood estimation (MLE) which maximizes the joint probability of training data. However, the most natural way of training would be finding the parameters that directly minimize the error rate of a given training set. In this article, we propose a novel learning method that minimizes the number of incorrectly decoded labels frame-wise. To do this, we construct a smooth function that is arbitrarily close to the exact frame error rate and minimize it directly using a gradient-based optimization algorithm. The proposed approach is intuitive and simple. We applied our method to the task of chord recognition in music, and the results show that it performs better than Maximum Likelihood Estimation and Minimum Classification Error.

Original languageEnglish (US)
Title of host publicationProceedings - 2012 11th International Conference on Machine Learning and Applications, ICMLA 2012
Pages363-368
Number of pages6
Volume2
DOIs
StatePublished - 2012
Event11th IEEE International Conference on Machine Learning and Applications, ICMLA 2012 - Boca Raton, FL, United States
Duration: Dec 12 2012Dec 15 2012

Other

Other11th IEEE International Conference on Machine Learning and Applications, ICMLA 2012
CountryUnited States
CityBoca Raton, FL
Period12/12/1212/15/12

Fingerprint

Hidden Markov models
Maximum likelihood estimation
learning method
Supervised learning
Labels
music

Keywords

  • chord recognition
  • HMM
  • machine learning
  • MFE
  • MLE

ASJC Scopus subject areas

  • Human-Computer Interaction
  • Education

Cite this

Cho, T., Kim, K., & Bello, J. P. (2012). A minimum frame error criterion for hidden markov model training. In Proceedings - 2012 11th International Conference on Machine Learning and Applications, ICMLA 2012 (Vol. 2, pp. 363-368). [6406763] https://doi.org/10.1109/ICMLA.2012.147

A minimum frame error criterion for hidden markov model training. / Cho, Taemin; Kim, Kibeom; Bello, Juan P.

Proceedings - 2012 11th International Conference on Machine Learning and Applications, ICMLA 2012. Vol. 2 2012. p. 363-368 6406763.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Cho, T, Kim, K & Bello, JP 2012, A minimum frame error criterion for hidden markov model training. in Proceedings - 2012 11th International Conference on Machine Learning and Applications, ICMLA 2012. vol. 2, 6406763, pp. 363-368, 11th IEEE International Conference on Machine Learning and Applications, ICMLA 2012, Boca Raton, FL, United States, 12/12/12. https://doi.org/10.1109/ICMLA.2012.147
Cho T, Kim K, Bello JP. A minimum frame error criterion for hidden markov model training. In Proceedings - 2012 11th International Conference on Machine Learning and Applications, ICMLA 2012. Vol. 2. 2012. p. 363-368. 6406763 https://doi.org/10.1109/ICMLA.2012.147
Cho, Taemin ; Kim, Kibeom ; Bello, Juan P. / A minimum frame error criterion for hidden markov model training. Proceedings - 2012 11th International Conference on Machine Learning and Applications, ICMLA 2012. Vol. 2 2012. pp. 363-368
@inproceedings{3b91dd91979e4e109df8137bb5558fe4,
title = "A minimum frame error criterion for hidden markov model training",
abstract = "Hidden Markov models (HMM) have been widely studied and applied over decades. The standard supervised learning method for HMM is maximum likelihood estimation (MLE) which maximizes the joint probability of training data. However, the most natural way of training would be finding the parameters that directly minimize the error rate of a given training set. In this article, we propose a novel learning method that minimizes the number of incorrectly decoded labels frame-wise. To do this, we construct a smooth function that is arbitrarily close to the exact frame error rate and minimize it directly using a gradient-based optimization algorithm. The proposed approach is intuitive and simple. We applied our method to the task of chord recognition in music, and the results show that it performs better than Maximum Likelihood Estimation and Minimum Classification Error.",
keywords = "chord recognition, HMM, machine learning, MFE, MLE",
author = "Taemin Cho and Kibeom Kim and Bello, {Juan P.}",
year = "2012",
doi = "10.1109/ICMLA.2012.147",
language = "English (US)",
isbn = "9780769549132",
volume = "2",
pages = "363--368",
booktitle = "Proceedings - 2012 11th International Conference on Machine Learning and Applications, ICMLA 2012",

}

TY - GEN

T1 - A minimum frame error criterion for hidden markov model training

AU - Cho, Taemin

AU - Kim, Kibeom

AU - Bello, Juan P.

PY - 2012

Y1 - 2012

N2 - Hidden Markov models (HMM) have been widely studied and applied over decades. The standard supervised learning method for HMM is maximum likelihood estimation (MLE) which maximizes the joint probability of training data. However, the most natural way of training would be finding the parameters that directly minimize the error rate of a given training set. In this article, we propose a novel learning method that minimizes the number of incorrectly decoded labels frame-wise. To do this, we construct a smooth function that is arbitrarily close to the exact frame error rate and minimize it directly using a gradient-based optimization algorithm. The proposed approach is intuitive and simple. We applied our method to the task of chord recognition in music, and the results show that it performs better than Maximum Likelihood Estimation and Minimum Classification Error.

AB - Hidden Markov models (HMM) have been widely studied and applied over decades. The standard supervised learning method for HMM is maximum likelihood estimation (MLE) which maximizes the joint probability of training data. However, the most natural way of training would be finding the parameters that directly minimize the error rate of a given training set. In this article, we propose a novel learning method that minimizes the number of incorrectly decoded labels frame-wise. To do this, we construct a smooth function that is arbitrarily close to the exact frame error rate and minimize it directly using a gradient-based optimization algorithm. The proposed approach is intuitive and simple. We applied our method to the task of chord recognition in music, and the results show that it performs better than Maximum Likelihood Estimation and Minimum Classification Error.

KW - chord recognition

KW - HMM

KW - machine learning

KW - MFE

KW - MLE

UR - http://www.scopus.com/inward/record.url?scp=84873580275&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84873580275&partnerID=8YFLogxK

U2 - 10.1109/ICMLA.2012.147

DO - 10.1109/ICMLA.2012.147

M3 - Conference contribution

AN - SCOPUS:84873580275

SN - 9780769549132

VL - 2

SP - 363

EP - 368

BT - Proceedings - 2012 11th International Conference on Machine Learning and Applications, ICMLA 2012

ER -