The learning dynamics of a universal approximator

Ansgar H. L. West, David Saad, Ian T. Nabney, Michael C. Mozer (Editor), Thomas Petsche (Editor), Michael I. Jordan (Editor)

Research output: Contribution to journalArticlepeer-review


The learning properties of a universal approximator, a normalized committee machine with adjustable biases, are studied for on-line back-propagation learning. Within a statistical mechanics framework, numerical studies show that this model has features which do not exist in previously studied two-layer network models without adjustable biases, e.g., attractive suboptimal symmetric phases even for realizable cases and noiseless data.
Original languageEnglish
Pages (from-to)288-294
Number of pages7
JournalAdvances in Neural Information Processing Systems
Publication statusPublished - May 1997

Bibliographical note

Copyright of the Massachusetts Institute of Technology Press (MIT Press)


  • approximator
  • back-propagation
  • symmetric phases
  • realizable cases
  • noiseless data


Dive into the research topics of 'The learning dynamics of a universal approximator'. Together they form a unique fingerprint.

Cite this