jufier@daimi.aau.dk (Svend Jules Fjerdingstad) (12/05/90)
In connection with parallel implementations of neural networks it is often desirable to present several input patterns before a weight update is performed. However, apparently such infrequent weight updates may sometimes influence the rate of convergence and the ability to generalize. If you know of any articles dealing with this subject, please mail references. We are especially interested in hearing about results concerning feed-forward networks. Apparently, the following article deals with the problems of infrequent weight updates. Do you know how we may get a copy of this article, or, alternatively, do you know the email address of any of the authors. Please mail responses. - Haffner P., Waibel A., Sawai H., & Shikano K. (1988): Fast back-propagation learning methods for neural networks in speech. (Tech. Rep. No. TR-1-0058) Osaka, Japan: ATR Interpreting Telephony Research Laboratories. Thanks a lot. Svend. -- Svend Jules Fjerdingstad, jufier@daimi.aau.dk | "To love, Computer Science Department, University of Aarhus | and to learn." Ny Munkegade 116, DK-8000 Aarhus C, DENMARK |