Presentation 1998/3/20
A Modular Neural Network Architecture for Pattern Classification : II: Recursive Decomposing and Parallel Learning of Large-Scale Problems
Bao-Liang Lu, Masami Ito,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In our previous paper, we have proposed a divide-and-conquer modular neural network learning paradigm for pattern classification. In this paper, we show the power and effectiveness of this paradigm through a large-scale benchmark problem, the shuttle problem, which is difficult to be learned by plain multilayer perceptrons (MLPs). We demonstrate how to decompose the shuttle problem into a series of independent smaller and simpler two-class problems based on class relations among the training data. We also demonstrate how to learn each of the two-class problems by a network module in parallel and recombine the outputs of the trained modules into a solution to the original problem. The simulation results show that our modular network has several significant advantages over plain MLPs such as fast learning and good generalization performance.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Task decomposition / Module combination / Parallel learning / Modular network
Paper #
Date of Issue

Conference Information
Committee NC
Conference Date 1998/3/20(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) A Modular Neural Network Architecture for Pattern Classification : II: Recursive Decomposing and Parallel Learning of Large-Scale Problems
Sub Title (in English)
Keyword(1) Task decomposition
Keyword(2) Module combination
Keyword(3) Parallel learning
Keyword(4) Modular network
1st Author's Name Bao-Liang Lu
1st Author's Affiliation Bio-Mimetic Control Research Center The Institute of Physical and Chemical Research (RIKEN)()
2nd Author's Name Masami Ito
2nd Author's Affiliation Bio-Mimetic Control Research Center The Institute of Physical and Chemical Research (RIKEN)
Date 1998/3/20
Paper #
Volume (vol) vol.97
Number (no) 624
Page pp.pp.-
#Pages 8
Date of Issue