next up previous contents
Next: Structure Up: No Title Previous: Introduction

The Self-Organizing Map

The SOM algorithm resembles other Vector Quantization (VQ) algorithms, like k-means [2], and is closely related to principal curves [11]. The important distinction from VQ techniques is that the neurons are organized on a regular grid and along with the selected neuron also its neighbors are updated, whereby the SOM performs an ordering of the neurons. In this respect the SOM is a multidimensional scaling methods which project data from input space to a lower-dimensional output space [52]. In the SOM similar vectors in the input space are projected onto nearby neurons on the map.





Juha Vesanto
Tue May 27 12:40:37 EET DST 1997