Community Research and Development Information Service - CORDIS

Abstract

The weight updating formula of Kohonen's algorithm may be viewed as the derivative of an objective error function when the input data are generated by a discrete probability distribution. A simple appealing characterisation for the reference vectors layout at the end of the learning phase has been derived from this interpretation. This characterisation can be used for defining a convergence index for Kohonen's algorithm, and for proposing a batch version of it. A first set of experiments with the batch approach is discussed.

Additional information

Authors: VARFIS A, JRC Ispra (IT);VERSINO C, JRC Ispra (IT)
Bibliographic Reference: Paper presented: ESANN - European Symposium on Artificial Neural Networks, Bruxelles (BE), April 7-9, 1993
Availability: Available from (1) as Paper EN 37405 ORA
Record Number: 199310577 / Last updated on: 1994-11-29
Category: PUBLICATION
Original language: en
Available languages: en