Relational Neural Gas

Paaßen B. Relational Neural Gas. Bielefeld University; 2018.This is a Java 7, fully MATLAB (R) compatible implementation of _Relational neural gas_. Relational neural gas is a clustering algorithm for distance data, meaning that you put in a matrix of pairwise distances D as well as a number of clusters K and you receive a distribution of your data points into K distinct clusters. It has first been proposed by [Hammer and Hasenfuss (2007)](https://doi.org/10.1007/978-3-540-74565-5_16) and is an extension of the _neural gas_ algorithm by [Martinetz and Schulten (1991)](https://www.ks.uiuc.edu/Publications/Papers/PDF/MART91B/MART91B.pdf). The basic idea of neural gas is to represent each cluster k in terms of a prototype wk which is responsible for all data points for which this prototype is the closest prototype. This prototype-based representation has two main advantages compared to other clustering methods: 1. It permits to extend the clustering to new data points by assigning new data points to their closest prototype (out-of-sample extension). 2. It permits to inspect the data set in terms of representative data points (i.e. the prototypes)