PARAMETER ESTIMATION USING BAYESIAN SEQUENTIAL LEARNING FOR LATENT CLASS MODELS

• Kuroda Masahiro

  Department of Socio-Information, Okayama University of Science

Bibliographic Information

Other Title

• ベイズ逐次学習による潜在クラスモデルのパラメータ推定
• ベイズ チクジ ガクシュウ ニ ヨル センザイ クラス モデル ノ パラメータ スイテイ

Abstract

In this paper, we discuss Bayesian sequential learning on the parameter estimation for latent class models. We give the hyper Dirichlet prior distribution with hyper parameters expressing the latent class model structure assuming local independence. In latent class models, the observations for manifest variables are obtained but the observations for latent variables are always missing, so that the posterior distribution has the mixture hyper Dirichlet distribution. We show the exact posterior distribution. Then, for the mixture posterior, the parameters can not express local independence structure but a conjugate for the prior distribution is also lost. In order to preserve these properties, we find an approximate distribution with the same means and the same average variance of the mixture posterior.

Journal

• Bulletin of the Computational Statistics of Japan

  Bulletin of the Computational Statistics of Japan 17 (1), 9-20, 2005

  Japanese Society of Computational Statistics

Keywords

• Local Independence
• Hyper Dirichlet Distribution
• Mixture Distribution
• Approximate Distribution

Details

• CRID: 1390001204381043712
• NII Article ID: 110001262161
• NII Book ID: AN10195854
• DOI: 10.20551/jscswabun.17.1_9
• ISSN: 21899789
• ISSN: 09148930
• NDL BIB ID: 7361262
• Web Site: http://id.ndl.go.jp/bib/7361262
• Web Site: https://ndlsearch.ndl.go.jp/books/R000000004-I7361262
• Text Lang: ja
• Data Source:
  • JaLC
  • NDL
  • CiNii Articles
• Abstract License Flag: Disallowed