Presentation 1998/5/22
A Construction of Feedforward Neural Networks for Large Scale Parallel Processing
Hitoshi Fujiwara, Itsuo Takanami,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In order that communication overhead and fan-in/out of neurons do not increase as the scale of multi-layered neural networks (MNN) becomes larger, the locally connected MNN in which each neuron has only three connected neighbors in terms of fan-in or fan-out has already been proposed where neurons in each layer are laid in one dimension and the number of layers becomes O(N)for the number of inputs N. In this paper, first, we show that the recognition ability of MNN remarkably decreases as the number of layers increases beyond a certain one. Hence, we propose an MNN in which each neuron has five connected neighbors in terms of fan-in or fan-out and neurons in each layer are laid in two dimension. Then we show that the number of layers becomes O(√). Further, we investigate the perfomances of the MNNs in which each neuron has some fixed connected neighbors, and neurons are connected in torus-like, respectively.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) multi-layered neural network / locally connected / large-scale system / parallel processing
Paper #
Date of Issue

Conference Information
Committee ICD
Conference Date 1998/5/22(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 Integrated Circuits and Devices (ICD)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) A Construction of Feedforward Neural Networks for Large Scale Parallel Processing
Sub Title (in English)
Keyword(1) multi-layered neural network
Keyword(2) locally connected
Keyword(3) large-scale system
Keyword(4) parallel processing
1st Author's Name Hitoshi Fujiwara
1st Author's Affiliation Matsushita Communication Sendai R & D Labs.Co., Ltd.()
2nd Author's Name Itsuo Takanami
2nd Author's Affiliation Department of Computer and Information Science, Iwate University
Date 1998/5/22
Paper #
Volume (vol) vol.98
Number (no) 66
Page pp.pp.-
#Pages 8
Date of Issue