CONNECT, Computational Neural Network Center

  • Hansen, Lars Kai (Project Manager)
  • Larsen, Jan (Project Participant)
  • Goutte, Cyril (Project Participant)
  • Ohlsson, Børje Ola Mattias (Project Participant)
  • Toft, Peter Aundal (Project Participant)
  • Fog, Torben L. (Project Participant)
  • Mørch, Niels J.S. (Project Participant)
  • Pedersen, Morten With (Project Participant)
  • Hintz-Madsen, Mads (Project Participant)
  • Kjems, Ulrik (Project Participant)
  • Nielsen, Johannes Kristoffer (Project Participant)
  • Lautrup, Benny (Project Participant)
  • Solla, Sara (Project Participant)
  • Winter, Ole (Project Participant)

    Project Details

    Description

    The Computational Neural Network Center was established March 1, 1991. The center's main research objective is actively to promote and support the collaboration between Danish researchers in theory, implementation and application of neural computation.
    An additional objective is to establish
    a graduate level training in the subject of artificial neural networks.
    In 1993 a plan funded by the Danish Research councils extended CONNECT for the period 1994-1996. The research plan is now centered around two projects: a theory project at the Niels Bohr Institute and the neural signal processing project at the Technical University of Denmark.
    Neural networks form an attractive framework for development of
    non-linear signal processing systems.
    They allow for system specification by "example" and thereby avoid explicit
    modeling. Arbitrary transfer functions may be modeled and neural net programs are "born" parallel facilitating implementation on massively parallel hardware.
    Theoretical tools for studying learning dynamics and generalization
    have matured considerably. Generalization, i.e., the ability to perform well on data not seen during adaptation, is the key concept for
    network design and evaluation.
    The research in 1996 concerned design, evaluation and visualization
    of non-linear adaptive models. A novel criterion for network pruning
    based on the generalization theory was formulated. A method for fast approximate crossvalidation of adaptive models was developed, and applied to system identification.
    The first scheme for generalization based
    evaluation of unsupervised learning algorithms was published and applied to optimization of Principal Component Analysis and k-Means Clustering.
    The Boltzmann Machine Learning Rule was generalized and applied to parameter estimation in inhomogeneous Markov Fields. The generalized form of the Boltzmann Machine network becomes susceptible to the generic tools for design and evaluation previously developed.
    StatusActive
    Effective start/end date01/03/1991 → …