| 電子情報通信学会 研究会発表申込システム 講演論文 詳細 |
技報閲覧サービス [ログイン] 技報アーカイブ |
| トップに戻る | 前のページに戻る | [Japanese] / [English] |
| 講演抄録／キーワード | ||
| 講演名 | 2011-11-09 15:45 Kernel and Feature Search in Kernel PCA ○Alam Md. Ashad（SOKENDAI）・Kenji Fukumizu（ISM）  IBISML2011-49 |
|
| 抄録 | （和） | During the last decade, unsupervised learning has become an important application area of the kernel methods. While the choice of kernel is essential in kernel methods for favorable performance, well-founded methods for kernel choice in unsupervised learning have not yet been established. We proposes a method for choosing parameters, the kernel and the number of features, of the kernel principal component analysis (kernel PCA) which is one of the most powerful tools in unsupervised learning. While the cross-validation has been applied to the standard PCA for choosing the number of components in kernel PCA we have a different feature space for each kernel, which makes it difficult to apply the cross-validation on the feature spaces for seeking a kernel and the number features. The propose method chooses them through the cross-validation based on pre-images. This paper mainly focuses two examples of kernels, Gaussian radial basis function (RBF) kernel and polynomial kernel. We have made experiments on simulated data as well as real world problems. The results show that the proposed method successfully chooses an appropriate kernel and the number of features in kernel PCA in terms of visualization and classification errors on the principal components. |
| （英） | During the last decade, unsupervised learning has become an important application area of the kernel methods. While the choice of kernel is essential in kernel methods for favorable performance, well-founded methods for kernel choice in unsupervised learning have not yet been established. We proposes a method for choosing parameters, the kernel and the number of features, of the kernel principal component analysis (kernel PCA) which is one of the most powerful tools in unsupervised learning. While the cross-validation has been applied to the standard PCA for choosing the number of components in kernel PCA we have a different feature space for each kernel, which makes it difficult to apply the cross-validation on the feature spaces for seeking a kernel and the number features. The propose method chooses them through the cross-validation based on pre-images. This paper mainly focuses two examples of kernels, Gaussian radial basis function (RBF) kernel and polynomial kernel. We have made experiments on simulated data as well as real world problems. The results show that the proposed method successfully chooses an appropriate kernel and the number of features in kernel PCA in terms of visualization and classification errors on the principal components. |
|
| キーワード | （和） | Unsupervised Learning / Kernel PCA / Pre-Iamges / Leave-one-out Cross-Validation / / / / |
| （英） | Unsupervised Learning / Kernel PCA / Pre-Iamges / Leave-one-out Cross-Validation / / / / | |
| 文献情報 | 信学技報, vol. 111, no. 275, IBISML2011-49, pp. 47-56, 2011年11月. | |
| 資料番号 | IBISML2011-49 | |
| 発行日 | 2011-11-02 (IBISML) | |
| ISSN | Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 | |
| 著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します．(許諾番号：10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) | |
| PDFダウンロード | IBISML2011-49 |
|
| 研究会情報 | |
| 研究会 | IBISML |
| 開催期間 | 2011-11-09 - 2011-11-11 |
| 開催地（和） | 奈良女子大学 |
| 開催地（英） | Nara Womens Univ. |
| テーマ（和） | 第14回情報論的学習理論ワークショップ |
| テーマ（英） | The 14th IBIS workshop |
| 講演論文情報の詳細 | |
| 申込み研究会 | IBISML |
| 会議コード | 2011-11-IBISML |
| 本文の言語 | 英語 |
| タイトル（和） | |
| サブタイトル（和） | |
| タイトル（英） | Kernel and Feature Search in Kernel PCA |
| サブタイトル（英） | |
| キーワード(1)（和/英） | Unsupervised Learning / Unsupervised Learning |
| キーワード(2)（和/英） | Kernel PCA / Kernel PCA |
| キーワード(3)（和/英） | Pre-Iamges / Pre-Iamges |
| キーワード(4)（和/英） | Leave-one-out Cross-Validation / Leave-one-out Cross-Validation |
| キーワード(5)（和/英） | / |
| キーワード(6)（和/英） | / |
| キーワード(7)（和/英） | / |
| キーワード(8)（和/英） | / |
| 第1著者 氏名（和/英/ヨミ） | Alam Md. Ashad / Alam Md. Ashad / |
| 第1著者 所属（和/英） | The Graduate University for Advanced Studies (略称： 総研大) The Graduate University for Advanced Studies (略称： SOKENDAI) |
| 第2著者 氏名（和/英/ヨミ） | Kenji Fukumizu / Kenji Fukumizu / |
| 第2著者 所属（和/英） | The Instituted of Statistical Mathematics (略称： ISM) The Instituted of Statistical Mathematics (略称： ISM) |
| 講演者 | 第1著者 |
| 発表日時 | 2011-11-09 15:45:00 |
| 発表時間 | 180分 |
| 申込先研究会 | IBISML |
| 資料番号 | IBISML2011-49 |
| 巻番号（vol） | vol.111 |
| 号番号（no） | no.275 |
| ページ範囲 | pp.47-56 |
| ページ数 | 10 |
| 発行日 | 2011-11-02 (IBISML) |
[研究会発表申込システムのトップページに戻る]