Improving Performance of Self-Organising Maps with Distance Metric Learning Method

TL;DR

This paper explores how replacing the Euclidean distance with a learned Mahalanobis metric improves the classification accuracy of Self-Organising Maps across various real-world datasets.

Contribution

It introduces a distance metric learning approach using Large Margin Nearest Neighbour to enhance SOM performance in pattern recognition tasks.

Findings

Improved classification accuracy on multiple datasets

Effective separation of classes using Mahalanobis metric

Enhanced SOM interpretability with metric learning

Abstract

Self-Organising Maps (SOM) are Artificial Neural Networks used in Pattern Recognition tasks. Their major advantage over other architectures is human readability of a model. However, they often gain poorer accuracy. Mostly used metric in SOM is the Euclidean distance, which is not the best approach to some problems. In this paper, we study an impact of the metric change on the SOM's performance in classification problems. In order to change the metric of the SOM we applied a distance metric learning method, so-called 'Large Margin Nearest Neighbour'. It computes the Mahalanobis matrix, which assures small distance between nearest neighbour points from the same class and separation of points belonging to different classes by large margin. Results are presented on several real data sets, containing for example recognition of written digits, spoken letters or faces.