Experiments with the Back Propagation Algorithm: A Systematic Look at a Small Problem.
Abstract
A multi-layer perception is a type of feed neural network for which there exists a learning algorithm based on error back-propagation. Experimental results are presented of the use of this error back-propagation algorithm on a carefully selected, simple problem. The effects of varying the structure of the network, the number of hidden units, the size of the training set and the initial weight values have been investigated. A number of ways of analysing solutions to such a simple problem are demonstrated. Explanations of the observed behaviour are offered which may provide insights applicable to a range of problems.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 29, 1987
- Accession Number
- ADA189266
Entities
People
- J. S. Bridle
- M. D. Bedworth
Organizations
- Royal Signals and Radar Establishment