English  |  正體中文  |  简体中文  |  Items with full text/Total items : 27221/39064
Visitors : 2403207      Online Users : 80
RC Version 4.0 © Powered By DSPACE, MIT. Enhanced by NTU Library IR team.
Scope Adv. Search

Please use this identifier to cite or link to this item: http://ntour.ntou.edu.tw:8080/ir/handle/987654321/46188

Title: Diffusion network architectures for implementation of Gibbs samplers with applications to assignment problems
Authors: Pei-Yih Ting
Ronald A. Iltis
Contributors: 國立臺灣海洋大學:資訊工程學系
NTOU:Department of Computer Science and Engineering
Keywords: Analog circuits;Neurons;Nonlinear equations;Feedback amplifiers;Character generation;Circuit simulation;Computer architecture;Analog computers;Concurrent computing;Computational complexity
Date: 1994-07
Issue Date: 2018-05-07T07:03:52Z
Publisher: IEEE Transactions on Neural Networks
Abstract: Abstract:
In this paper, analog circuit designs for implementations of Gibbs samplers are presented, which offer fully parallel computation. The Gibbs sampler for a discrete solution space (or Boltzmann machine) can be used to solve both deterministic and probabilistic assignment (association) problems. The primary drawback to the use of a Boltzmann machine for optimization is its computational complexity, since updating of the neurons is typically performed sequentially. We first consider the diffusion equation emulation of a Boltzmann machine introduced by Roysam and Miller (1991), which employs a parallel network of nonlinear amplifiers. It is shown that an analog circuit implementation of the diffusion equation requires a complex neural structure incorporating matched nonlinear feedback amplifiers and current multipliers. We introduce a simpler implementation of the Boltzmann machine, using a "constant gradient" diffusion equation, which eliminates the need for a matched feedback amplifier. The performance of the Roysam and Miller network and the new constant gradient (CG) network is compared using simulations for the multiple-neuron case, and integration of the Chapman-Kolmogorov equation for a single neuron. Based on the simulation results, heuristic criteria for establishing the diffusion equation boundaries, and neuron sigmoidal gain are obtained. The final CG analog circuit is suitable for VLSI implementation, and hence may offer rapid convergence.
Relation: 5(4), pp.622-638
URI: http://ntour.ntou.edu.tw:8080/ir/handle/987654321/46188
Appears in Collections:[資訊工程學系] 期刊論文

Files in This Item:

File Description SizeFormat

All items in NTOUR are protected by copyright, with all rights reserved.


著作權政策宣告: 本網站之內容為國立臺灣海洋大學所收錄之機構典藏,無償提供學術研究與公眾教育等公益性使用,請合理使用本網站之內容,以尊重著作權人之權益。
網站維護: 海大圖資處 圖書系統組
DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library IR team Copyright ©   - Feedback