The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9707

Title:An Investigation of Adaptive Neural Networks
Date:Sep 1997
Abstract:Although much work has been done on allowing feedforward networks to grow in response to novel input data, less work has been done with self-organising maps. A problem that arises with an associative memory of fixed size is that the learning of new data may disrupt the categorisation of information already stored in the network. Either the net can be allowed to overwrite data which is not important, according to some set of criteria, or the net can be allowed to grow, or some combination of the two can be utilised to overcome the problem. This project conctntrates on how the growth of a self-organising feature map might be controlled. Some of the different models currently being explored are reviewed and two novel approaches are evaluated. Both were found to give good performance in modelling the distribution of input data.

[Search These Pages] [DAI Home Page] [Comment]