Design and Parallel Implementation of a Dynamic Node Growing Neural Network Construction Algorithm on Scalable Parallel Machines

スポンサーリンク

概要

• 論文の詳細を見る

• In this paper, we first present a novel method for constructing a feedforward neural network with a single hidden layer dynamically and then prove its convergence. Following that is a description of its parallel implementation on scalable parallel machines. The dynamic node growing method starts with a small number of hidden units and constructs the neural network by adding one hidden unit at a time until the network achieves a desired accuracy. At each stage of the construction process, an optimal set of weights for the growing network is obtained by applying a variant of the quasi-Newton method for unconstrained optimization. We propose two algorithms for parallel implementation of this method : sample-partition-based and hidden-unit-partition-based algorithms. Experimental results obtained from our implementation of these parallel algorithms indicate that a significant performance improvement can be achieved on scalable parallel machines.

• 社団法人人工知能学会の論文
• 1995-07-01

著者

• Gao Y

  Univ. Tokyo Tokyo Jpn

• Gao Yaoqing

  Dept. Of Information Science Faculty Of Science The University Of Tokyo:the Research Of This Author

• Setiono Rusy

  Dept. of Information Systems and Computer Science, National University of Singapore

• Setiono Rusy

  Dept. Of Information Systems And Computer Science National University Of Singapore

関連論文

• A Comparison of Parallel α-β Search Algorithms
• Design and Parallel Implementation of a Dynamic Node Growing Neural Network Construction Algorithm on Scalable Parallel Machines

スポンサーリンク