Abstract for prager_tr106

Cambridge University Engineering Department Technical Report CUED/F-INFENG/TR106

SOME EXPERIMENTS WITH FIXED NON-LINEAR MAPPINGS AND SINGLE LAYER NETWORKS

Richard W. Prager

July 1992

Kanerva's sparse distributed memory model consists of a fixed non-linear mapping, called location matching, followed by a single layer of adaptive dot-product step-threshold links. Various networks of this type are tested on three tasks in order to discover the circumstances in which this type of network provides an efficient solution.

The networks provide more competitive performance when the dimensionality of the input patterns is fairly low. For high dimensional inputs the performance of the single layer network alone is better on the test data, whereas for low dimensional inputs a two layer adaptive network is required and this needs much more training before it can produce a better test performance.

A new `location pruning' technique is reported which improves the design of the location matching mappings. The resulting network is extensively tested on large pattern classification tasks to demonstrate the benefits of the algorithm.

The experiments show that the main benefit of the location matching networks is their ability, using the location pruning algorithm, to train in roughly 1/2 to 1/10 of the training iterations required by single or double layer adaptive networks on the same tasks.


(ftp:) prager_tr106.ps.Z (http:) prager_tr106.ps.Z
PDF (automatically generated from original PostScript document - may be badly aliased on screen):
  (ftp:) prager_tr106.pdf | (http:) prager_tr106.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.