Using feature trimming to improve the performance of Dystal
David Glenn Clark · 2003
Dystal is a simple, biologically-based artificial neural network which trains much faster than backpropagation. It's developers use the correlation coefficient as a measure of similarity when using Dystal to solve image processing problems. The correlation coefficient is not suitable as a distance measure between points in general data sets. In such data sets the Mahalauobis distance is more appropriate. The performance of Dystal with the Mahalauobis distance can be improved by removing "noise" features from the data set.