Estimation of probability density functions from limited training samples

Chulhee Lee, Euisun Choi

Research output: Contribution to conferencePaper

Abstract

In this paper, we analyze probability density functions when the number of training samples is limited, assuming normal distributions. As the dimension of data increases significantly, the performance of a classifier suffers when the number of training samples is not adequate. This problem becomes worse as high dimensional data such as hyperspectral images are widely available. The key factor in designing a classifier is estimation of probability density functions, which are completely determined by covariance matrices and mean vectors in case of the Gaussian ML classifier. In this paper, we provide in-depth analyses of estimation of probability density functions in terms of the number of training samples assuming normal distributions and provide a guideline in choosing the dimensionality of data for a given set of training samples.

Original languageEnglish
Pages458-462
Number of pages5
Publication statusPublished - 2004 Dec 27
EventProceedings of the Seventh IASTED International Conference on Computer Graphics and Imaging - Kauai, HI, United States
Duration: 2004 Aug 172004 Aug 19

Other

OtherProceedings of the Seventh IASTED International Conference on Computer Graphics and Imaging
CountryUnited States
CityKauai, HI
Period04/8/1704/8/19

Fingerprint

Probability density function
Classifiers
Normal distribution
Covariance matrix

All Science Journal Classification (ASJC) codes

  • Engineering(all)

Cite this

Lee, C., & Choi, E. (2004). Estimation of probability density functions from limited training samples. 458-462. Paper presented at Proceedings of the Seventh IASTED International Conference on Computer Graphics and Imaging, Kauai, HI, United States.
Lee, Chulhee ; Choi, Euisun. / Estimation of probability density functions from limited training samples. Paper presented at Proceedings of the Seventh IASTED International Conference on Computer Graphics and Imaging, Kauai, HI, United States.5 p.
@conference{1b96d948561a45e3812cc219b3626fed,
title = "Estimation of probability density functions from limited training samples",
abstract = "In this paper, we analyze probability density functions when the number of training samples is limited, assuming normal distributions. As the dimension of data increases significantly, the performance of a classifier suffers when the number of training samples is not adequate. This problem becomes worse as high dimensional data such as hyperspectral images are widely available. The key factor in designing a classifier is estimation of probability density functions, which are completely determined by covariance matrices and mean vectors in case of the Gaussian ML classifier. In this paper, we provide in-depth analyses of estimation of probability density functions in terms of the number of training samples assuming normal distributions and provide a guideline in choosing the dimensionality of data for a given set of training samples.",
author = "Chulhee Lee and Euisun Choi",
year = "2004",
month = "12",
day = "27",
language = "English",
pages = "458--462",
note = "Proceedings of the Seventh IASTED International Conference on Computer Graphics and Imaging ; Conference date: 17-08-2004 Through 19-08-2004",

}

Lee, C & Choi, E 2004, 'Estimation of probability density functions from limited training samples' Paper presented at Proceedings of the Seventh IASTED International Conference on Computer Graphics and Imaging, Kauai, HI, United States, 04/8/17 - 04/8/19, pp. 458-462.

Estimation of probability density functions from limited training samples. / Lee, Chulhee; Choi, Euisun.

2004. 458-462 Paper presented at Proceedings of the Seventh IASTED International Conference on Computer Graphics and Imaging, Kauai, HI, United States.

Research output: Contribution to conferencePaper

TY - CONF

T1 - Estimation of probability density functions from limited training samples

AU - Lee, Chulhee

AU - Choi, Euisun

PY - 2004/12/27

Y1 - 2004/12/27

N2 - In this paper, we analyze probability density functions when the number of training samples is limited, assuming normal distributions. As the dimension of data increases significantly, the performance of a classifier suffers when the number of training samples is not adequate. This problem becomes worse as high dimensional data such as hyperspectral images are widely available. The key factor in designing a classifier is estimation of probability density functions, which are completely determined by covariance matrices and mean vectors in case of the Gaussian ML classifier. In this paper, we provide in-depth analyses of estimation of probability density functions in terms of the number of training samples assuming normal distributions and provide a guideline in choosing the dimensionality of data for a given set of training samples.

AB - In this paper, we analyze probability density functions when the number of training samples is limited, assuming normal distributions. As the dimension of data increases significantly, the performance of a classifier suffers when the number of training samples is not adequate. This problem becomes worse as high dimensional data such as hyperspectral images are widely available. The key factor in designing a classifier is estimation of probability density functions, which are completely determined by covariance matrices and mean vectors in case of the Gaussian ML classifier. In this paper, we provide in-depth analyses of estimation of probability density functions in terms of the number of training samples assuming normal distributions and provide a guideline in choosing the dimensionality of data for a given set of training samples.

UR - http://www.scopus.com/inward/record.url?scp=10444277337&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=10444277337&partnerID=8YFLogxK

M3 - Paper

AN - SCOPUS:10444277337

SP - 458

EP - 462

ER -

Lee C, Choi E. Estimation of probability density functions from limited training samples. 2004. Paper presented at Proceedings of the Seventh IASTED International Conference on Computer Graphics and Imaging, Kauai, HI, United States.