Abstract for burrows_tr158

Cambridge University Engineering Department Technical Report CUED/F-INFENG/TR158

FEED-FORWARD AND RECURRENT NEURAL NETWORKS FOR SYSTEM IDENTIFICATION

T. L. Burrows and M. Niranjan

December 1993

Neural networks with nonlinear sigmoid functions at the hidden nodes have been shown to give improved performance over linear models for time series prediction problems. We demonstrate that whenever the network produces a useful solution to this problem, the hidden nodes operate predominantly in the linear region of their sigmoid function, and that small excursions into the nonlinear region allow improved prediction. Using several nonlinear time series, we demonstrate that this allows us to exploit standard linear system identification techniques. For a speech prediction problem, we compare the performance of a feed-forward and recurrent architecture and in view of our observations, attribute the improved performance of the recurrent network to a parameter estimation based on output error minimisation, rather than the equation error minimisation performed by feed-forward networks and linear prediction analysis.


(ftp:) burrows_tr158.ps.Z (http:) burrows_tr158.ps.Z
PDF (automatically generated from original PostScript document - may be badly aliased on screen):
  (ftp:) burrows_tr158.pdf | (http:) burrows_tr158.pdf

If you have difficulty viewing files that end '.gz', which are gzip compressed, then you may be able to find tools to uncompress them at the gzip web site.

If you have difficulty viewing files that are in PostScript, (ending '.ps' or '.ps.gz'), then you may be able to find tools to view them at the gsview web site.

We have attempted to provide automatically generated PDF copies of documents for which only PostScript versions have previously been available. These are clearly marked in the database - due to the nature of the automatic conversion process, they are likely to be badly aliased when viewed at default resolution on screen by acroread.