Phys. Rev. Lett. 78, 555 - 558 (1997)Finite Size Scaling in Neural Networks |
PRL Celebrates 50 Years
This Week's Milestone Letters are from 1984: |
Walter Nadler
Institut für Theoretische Chemie, Universität Tübingen, Auf der Morgenstelle 8, D-72076 Tübingen, Germany
Wolfgang Fink
Institut für Theoretische Physik, Universität Tübingen, Auf der Morgenstelle 14, D-72076 Tübingen, Germany
Received 25 July 1996
We demonstrate that the fraction of pattern sets that can be stored in single- and hidden-layer perceptrons exhibits finite size scaling. This feature allows one to estimate the critical storage capacity αc from simulations of relatively small systems. We illustrate this approach by determining αc, together with the finite size scaling exponent ν, for storing Gaussian patterns in committee and parity machines with binary couplings and up to K = 5 hidden units.
©1997 The American Physical Society
URL: http://link.aps.org/abstract/PRL/v78/p555
DOI: 10.1103/PhysRevLett.78.555
PACS: 87.10.+e, 02.70.Lq, 05.50.+q, 64.60.Cn
See Also
Comment: M. Schröder and R. Urbanczik, Comment on “Finite Size Scaling in Neural Networks”, Phys. Rev. Lett. 80, 4109 (1998)
[ Abstract | Previous article | Next article | Issue 3 ]
About PROLA | Terms and Conditions | Subscriptions | Search | Help
Content in PROLA © 2008 by The American Physical Society All rights reserved. PROLA™ is a trademark of The American Physical Society. Use of APS online journals implies that the user has read and agrees to the Terms and Conditions in the Subscription Agreement.