A Method for Task Allocation in Modular Neural Network with an Information Criterion

  • H.-H. Kim
  • Y. Anzai
Conference paper


It is well known that large-scale neural networks suffer from serious problems such as the scale problem and the local minima problems. Modular architecture neural network is an approach to alleviate these problems. It is important that the construction of modular neural network is the selection or construction of a network that can converge and has the good generalization ability for a task, and the Akaike Information Criterion (AIC) is a criterion of evaluation of estimated model from observed parameters is a very useful tool for selection of network. This paper proposes a method for task allocation in a modular architecture neural network. The method allocates a best fit network that has a good generalization ability from multiple neural networks for a task with (AIC) and the state of convergence of a network, simply and certainly. The performance of proposed method is evaluated with the Fisher’ Iris data and the What and Where vision tasks.


Akaike Information Criterion Hide Node Task Allocation Hide Unit Output Unit 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. [1]
    B.L.M. Happel and J.M.J. Murre, ‘Design and Evolution of Modular Neural Network Architectures’, Neural Networks, Volume 7, pages 985–1004, 1994.CrossRefGoogle Scholar
  2. [2]
    R.A. Jacobs, M.I. Jordan and A. G. Barto, ‘Task Decomposition Through Competition in a Modular Connectionist Architecture: The What and Where Vision Tasks’, Cognitive Science 15, pages 219–250, 1991.CrossRefGoogle Scholar
  3. [3]
    R.A. Jacobs and M.I. Jordan, ‘A Competitive Modular Connectionist Architecture’, Neural Information Processing Systems, Volume 3, pages 767–773, 1991.Google Scholar
  4. [4]
    S.J. Nowlan and G.E. Hinton, ‘Evaluation of Adaptive Mixtures of Competing Experts’, Neural Information Processing Systems, Volume 3, pages 774–780, 1991.Google Scholar
  5. [5]
    E. Ronco and P. Gawthrop, ‘Modular Neural Networks: A State of the Art’, CSC-95026, University of Glasgow, UK, 1995.Google Scholar
  6. [6]
    Y. Sakamoto, M. Ishiguro and G. Kitagawa, ‘Akaike Information Criterion Statistics’, D. Reidel Publishing Company, 1986.Google Scholar

Copyright information

© Springer-Verlag Wien 1998

Authors and Affiliations

  • H.-H. Kim
    • 1
  • Y. Anzai
    • 1
  1. 1.Dept. of Computer ScienceKeio UniversityKohoku-ku, Yokohama 223Japan

Personalised recommendations