GPD-based state modification by weighted linear loss function

Taehee Kwon, Hanseok Ko

Research output: Contribution to journalArticle

Abstract

This paper proposes an effective loss function to assign the HMM state weight based on the MCE method. This is to remedy the performance limitation inherent in the traditional maximum likelihood method, which adjusts parameters to maximize the likelihood of training HMM. If minimum classification error method is used to minimize the error rate for training data, then the local optimum point can be achieved and the recognition performance can be achieved. However, if the amount of data used in the MCE training is too small, there can be a risk of overfilling to the training data. In this paper, we propose state modification by weighted linear loss function to overcome overfilling to training data. Representative experiments confirm this postulation and show the improvement in error rate when applied.

Original languageEnglish
Pages (from-to)1100-1108
Number of pages9
JournalLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume2660
Publication statusPublished - 2003 Dec 1

Fingerprint

Loss Function
Linear Function
Error Rate
Maximum likelihood
Maximum Likelihood Method
Weights and Measures
Assign
Likelihood
Maximise
Training
Minimise
Experiments
Experiment

ASJC Scopus subject areas

  • Biochemistry, Genetics and Molecular Biology(all)
  • Computer Science(all)
  • Theoretical Computer Science

Cite this

@article{a62898926e7c48a892c30ce5b9eddccb,
title = "GPD-based state modification by weighted linear loss function",
abstract = "This paper proposes an effective loss function to assign the HMM state weight based on the MCE method. This is to remedy the performance limitation inherent in the traditional maximum likelihood method, which adjusts parameters to maximize the likelihood of training HMM. If minimum classification error method is used to minimize the error rate for training data, then the local optimum point can be achieved and the recognition performance can be achieved. However, if the amount of data used in the MCE training is too small, there can be a risk of overfilling to the training data. In this paper, we propose state modification by weighted linear loss function to overcome overfilling to training data. Representative experiments confirm this postulation and show the improvement in error rate when applied.",
author = "Taehee Kwon and Hanseok Ko",
year = "2003",
month = "12",
day = "1",
language = "English",
volume = "2660",
pages = "1100--1108",
journal = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
issn = "0302-9743",
publisher = "Springer Verlag",

}

TY - JOUR

T1 - GPD-based state modification by weighted linear loss function

AU - Kwon, Taehee

AU - Ko, Hanseok

PY - 2003/12/1

Y1 - 2003/12/1

N2 - This paper proposes an effective loss function to assign the HMM state weight based on the MCE method. This is to remedy the performance limitation inherent in the traditional maximum likelihood method, which adjusts parameters to maximize the likelihood of training HMM. If minimum classification error method is used to minimize the error rate for training data, then the local optimum point can be achieved and the recognition performance can be achieved. However, if the amount of data used in the MCE training is too small, there can be a risk of overfilling to the training data. In this paper, we propose state modification by weighted linear loss function to overcome overfilling to training data. Representative experiments confirm this postulation and show the improvement in error rate when applied.

AB - This paper proposes an effective loss function to assign the HMM state weight based on the MCE method. This is to remedy the performance limitation inherent in the traditional maximum likelihood method, which adjusts parameters to maximize the likelihood of training HMM. If minimum classification error method is used to minimize the error rate for training data, then the local optimum point can be achieved and the recognition performance can be achieved. However, if the amount of data used in the MCE training is too small, there can be a risk of overfilling to the training data. In this paper, we propose state modification by weighted linear loss function to overcome overfilling to training data. Representative experiments confirm this postulation and show the improvement in error rate when applied.

UR - http://www.scopus.com/inward/record.url?scp=35248828600&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=35248828600&partnerID=8YFLogxK

M3 - Article

AN - SCOPUS:35248828600

VL - 2660

SP - 1100

EP - 1108

JO - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

JF - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SN - 0302-9743

ER -