Presentation 2003/7/22
[Tutorial] Ensemble learning
Masato OKADA, Kazuyuki HARA, Seiji MIYOSHI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) We discuss emsemble learning algorithms, i. e., bagging and parallel boosting based on the on-line learning from statistical mechnical point of view. We treat cases that both of teacher and students are simple perceptrons. We show the the generalization error of emsemble learnig machine depends only on the normls of the weight vectors of student perceptrons, overlaps (direction cosine) between the weight vector of the teacher perceptron and those of the student perceptrons, and correlations between the weight vectors of the student percetrons. We derive differential equations to describe the dynamics of these macrosocpic variables. Applying these equations to a linear perceptron case and a nonlinear perceptron case, we discuss properties of the bagging and the parallel boosting in these cases.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Ensemble learning / Bagging / Parallel boosting / On-line learning / Statistical mechanics / Perceptron
Paper # NC2003-35
Date of Issue

Conference Information
Committee NC
Conference Date 2003/7/22(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) [Tutorial] Ensemble learning
Sub Title (in English)
Keyword(1) Ensemble learning
Keyword(2) Bagging
Keyword(3) Parallel boosting
Keyword(4) On-line learning
Keyword(5) Statistical mechanics
Keyword(6) Perceptron
1st Author's Name Masato OKADA
1st Author's Affiliation RIKEN Brain Science Institute:"Intelligent Cooperation and Control" PRESTO, JST()
2nd Author's Name Kazuyuki HARA
2nd Author's Affiliation Tokyo Metropolitan College of Technology
3rd Author's Name Seiji MIYOSHI
3rd Author's Affiliation Kobe City College of Technology
Date 2003/7/22
Paper # NC2003-35
Volume (vol) vol.103
Number (no) 228
Page pp.pp.-
#Pages 6
Date of Issue