A new maximum likelihood gradient algorithm for on-line hidden Markov model identification

Lain B. Collings; Tobias Ryden

doi:10.1109/ICASSP.1998.681599

A new maximum likelihood gradient algorithm for on-line hidden Markov model identification

Lain B. Collings, Tobias Ryden

Research output: Chapter in Book/Report/Conference proceeding › Conference proceeding contribution › peer-review

18 Citations (Scopus)

Abstract

This paper presents a new algorithm for on-line identification of hidden Markov model (HMM) parameters. The scheme is gradient based, and provides parameter estimates which recursively maximise the likelihood function. It is therefore a recursive maximum likelihood (RML) algorithm, and it has optimal asymptotic properties. The only current on-line HMM identification algorithm with anything other than suboptimal rate of convergence is based on a prediction error (PE) cost function. As well as presenting a new algorithm, this paper also highlights and explains a counter-intuitive convergence problem for the current recursive PE (RPE) algorithm, when operating in low noise conditions. Importantly, this problem does not exist for the new RML algorithm. Simulation studies demonstrate the superior performance of the new algorithm. compared to current techniques.

Original language	English
Title of host publication	Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 1998
Place of Publication	Piscataway, N.J.
Publisher	Institute of Electrical and Electronics Engineers (IEEE)
Pages	2261-2264
Number of pages	4
Volume	6
ISBN (Electronic)	0780344308, 0780344316
ISBN (Print)	0780344286, 0780344294
DOIs	https://doi.org/10.1109/ICASSP.1998.681599
Publication status	Published - May 1998
Externally published	Yes
Event	1998 23rd IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP - 1998 - Seattle, United States Duration: 12 May 1998 → 15 May 1998

Other

Other	1998 23rd IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP - 1998
Country/Territory	United States
City	Seattle
Period	12/05/98 → 15/05/98

Access to Document

10.1109/ICASSP.1998.681599

Cite this

Collings, L. B., & Ryden, T. (1998). A new maximum likelihood gradient algorithm for on-line hidden Markov model identification. In Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 1998 (Vol. 6, pp. 2261-2264). Article 681599 Institute of Electrical and Electronics Engineers (IEEE). https://doi.org/10.1109/ICASSP.1998.681599

@inproceedings{895e0113ac704bf1a2280e4037c4f3c8,

title = "A new maximum likelihood gradient algorithm for on-line hidden Markov model identification",

abstract = "This paper presents a new algorithm for on-line identification of hidden Markov model (HMM) parameters. The scheme is gradient based, and provides parameter estimates which recursively maximise the likelihood function. It is therefore a recursive maximum likelihood (RML) algorithm, and it has optimal asymptotic properties. The only current on-line HMM identification algorithm with anything other than suboptimal rate of convergence is based on a prediction error (PE) cost function. As well as presenting a new algorithm, this paper also highlights and explains a counter-intuitive convergence problem for the current recursive PE (RPE) algorithm, when operating in low noise conditions. Importantly, this problem does not exist for the new RML algorithm. Simulation studies demonstrate the superior performance of the new algorithm. compared to current techniques.",

author = "Collings, {Lain B.} and Tobias Ryden",

year = "1998",

month = may,

doi = "10.1109/ICASSP.1998.681599",

language = "English",

isbn = "0780344286",

volume = "6",

pages = "2261--2264",

booktitle = "Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 1998",

publisher = "Institute of Electrical and Electronics Engineers (IEEE)",

address = "United States",

note = "1998 23rd IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP - 1998 ; Conference date: 12-05-1998 Through 15-05-1998",

}

Collings, LB & Ryden, T 1998, A new maximum likelihood gradient algorithm for on-line hidden Markov model identification. in Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 1998. vol. 6, 681599, Institute of Electrical and Electronics Engineers (IEEE), Piscataway, N.J., pp. 2261-2264, 1998 23rd IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP - 1998, Seattle, Washington, United States, 12/05/98. https://doi.org/10.1109/ICASSP.1998.681599

A new maximum likelihood gradient algorithm for on-line hidden Markov model identification. / Collings, Lain B.; Ryden, Tobias.
Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 1998. Vol. 6 Piscataway, N.J.: Institute of Electrical and Electronics Engineers (IEEE), 1998. p. 2261-2264 681599.

Research output: Chapter in Book/Report/Conference proceeding › Conference proceeding contribution › peer-review

TY - GEN

T1 - A new maximum likelihood gradient algorithm for on-line hidden Markov model identification

AU - Collings, Lain B.

AU - Ryden, Tobias

PY - 1998/5

Y1 - 1998/5

N2 - This paper presents a new algorithm for on-line identification of hidden Markov model (HMM) parameters. The scheme is gradient based, and provides parameter estimates which recursively maximise the likelihood function. It is therefore a recursive maximum likelihood (RML) algorithm, and it has optimal asymptotic properties. The only current on-line HMM identification algorithm with anything other than suboptimal rate of convergence is based on a prediction error (PE) cost function. As well as presenting a new algorithm, this paper also highlights and explains a counter-intuitive convergence problem for the current recursive PE (RPE) algorithm, when operating in low noise conditions. Importantly, this problem does not exist for the new RML algorithm. Simulation studies demonstrate the superior performance of the new algorithm. compared to current techniques.

AB - This paper presents a new algorithm for on-line identification of hidden Markov model (HMM) parameters. The scheme is gradient based, and provides parameter estimates which recursively maximise the likelihood function. It is therefore a recursive maximum likelihood (RML) algorithm, and it has optimal asymptotic properties. The only current on-line HMM identification algorithm with anything other than suboptimal rate of convergence is based on a prediction error (PE) cost function. As well as presenting a new algorithm, this paper also highlights and explains a counter-intuitive convergence problem for the current recursive PE (RPE) algorithm, when operating in low noise conditions. Importantly, this problem does not exist for the new RML algorithm. Simulation studies demonstrate the superior performance of the new algorithm. compared to current techniques.

UR - http://www.scopus.com/inward/record.url?scp=0031632654&partnerID=8YFLogxK

U2 - 10.1109/ICASSP.1998.681599

DO - 10.1109/ICASSP.1998.681599

M3 - Conference proceeding contribution

AN - SCOPUS:0031632654

SN - 0780344286

SN - 0780344294

VL - 6

SP - 2261

EP - 2264

BT - Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 1998

PB - Institute of Electrical and Electronics Engineers (IEEE)

CY - Piscataway, N.J.

T2 - 1998 23rd IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP - 1998

Y2 - 12 May 1998 through 15 May 1998

ER -

Collings LB, Ryden T. A new maximum likelihood gradient algorithm for on-line hidden Markov model identification. In Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 1998. Vol. 6. Piscataway, N.J.: Institute of Electrical and Electronics Engineers (IEEE). 1998. p. 2261-2264. 681599 doi: 10.1109/ICASSP.1998.681599

A new maximum likelihood gradient algorithm for on-line hidden Markov model identification

Abstract

Other

Access to Document

Other files and links

Fingerprint

Cite this