Bayesian multiple and co-clustering methods: Application to fMRI data

• Tomoki Tokuda

  Okinawa Institute of Science and Technology Graduate University

• Junichiro Yoshimoto

  Okinawa Institute of Science and Technology Graduate University

• Yu Shimizu

  Okinawa Institute of Science and Technology Graduate University

• Kosuke Yoshida

  Kyoto University|Okinawa Institute of Science and Technology Graduate University

• Shigeru Toki

  Hiroshima University

• Go Okada

  Hiroshima University

• Masahiro Takamura

  Hiroshima University

• Tetsuya Yamamoto

  Hiroshima University

• Shinpei Yoshimura

  Otemon Gakuin University

• Yasumasa Okamoto

  Hiroshima University

• Shigeto Yamawaki

  Hiroshima University

• Noriaki Yahata

  Tokyo University

• Kenji Doya

  Okinawa Institute of Science and Technology Graduate University

抄録

We propose a novel approach for the dimension reduction of high dimensional data to make the data available for conventional statistical evaluations. Our method is based on nonparametric multiple Gaussian clustering, in which we assume that in each cluster block, the instances follow an independent and identically (i.i.d.) univariate Gaussian distribution. We show theoretically that our model can fit multivariate Gaussian distributions with exchangeable features. We further show how the clusters derived with this specific model can be used to effectively reduce the dimension of data taking into account associations between attributes. Finally, we demonstrate our approach in an application to resting state functional magnetic resonance imaging (fMRI) data, which implies subtypes of depression may be characterized by the treatment effect of antidepressant drug SSRI.

収録刊行物

• 情報処理学会研究報告. ICS, [知能と複雑系]

  情報処理学会研究報告. ICS, [知能と複雑系] 2014 (2), 1-5, 2014-07-15

  一般社団法人情報処理学会

詳細情報

• CRID: 1570572702898173440
• NII論文ID: 110009804782
• NII書誌ID: AA11135936
• ISSN: 09196072
• 本文言語コード: en
• データソース種別:
  • CiNii Articles