Presentation 2005-02-24
MDL-based nonlinear regression tree
Yoshimi UEZU, Takayuki NAKAMURA, Toshikazu WADA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In the previous work, we proposed the "PaLM-Tree" which is a kind of linear regression tree. It improves some drawbacks which the conventional linear regression tree method have. When it is applied for estimating a complex nonlinear mapping, it splits the input space into too much regions. In such case, the generalization performance of the PaLM-tree tends to be worse. In order to improve the generalization performance of the PaLM-tree, we proposed a MDL-based nonlinear regression tree. Instead of using linear regression functions, this method uses polynomial functions. Furthermore, it utilizes MDL criterion to decide the degree of polynomial and the splitting points. Through the experiments on function estimation problems including artifial data and actual data, we confirmed the advantages of the proposed method.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) PaLM-tree / Nonlinear regression tree / MDL criterion
Paper # NLC2004-99,PRMU2004-181
Date of Issue

Conference Information
Committee NLC
Conference Date 2005/2/17(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 Natural Language Understanding and Models of Communication (NLC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) MDL-based nonlinear regression tree
Sub Title (in English)
Keyword(1) PaLM-tree
Keyword(2) Nonlinear regression tree
Keyword(3) MDL criterion
1st Author's Name Yoshimi UEZU
1st Author's Affiliation Department of Computer and Communication Science, Wakayama University()
2nd Author's Name Takayuki NAKAMURA
2nd Author's Affiliation Department of Computer and Communication Science, Wakayama University
3rd Author's Name Toshikazu WADA
3rd Author's Affiliation Department of Computer and Communication Science, Wakayama University
Date 2005-02-24
Paper # NLC2004-99,PRMU2004-181
Volume (vol) vol.104
Number (no) 667
Page pp.pp.-
#Pages 6
Date of Issue