Saarland University, Machine Learning Group, Fak. MI - Mathematik und Informatik, Campus E1 1, 66123 Saarbr├╝cken, Germany     

Machine Learning Group
Department of Mathematics and Computer Science - Saarland University


a Matlab GUI to explore spectral clustering and the influence of different similarity graphs

by Matthias Hein and Ulrike von Luxburg


DemoSpectralClustering: In this demo, we would like to show how (normalized) spectral clustering behaves for different kinds of neighborhood graphs. In particular, we want to explore how the first eigenvectors of the graph Laplacian and the resulting low-dimensional embeddings react for different choices of neighborhood graphs.


DemoSpectralClustering has been used for teaching purposes at the Machine Learning Summer School 2007 in Tuebingen, Germany. The tutorial which is based on DemoSpectralClustering introduces the theoretical foundations of spectral clustering as well as given practical hints for spectral clustering using behavior on toy datasets.

Download tutorial on spectral clustering   

SCREENSHOT OF DemoSimilarityGraphs

PANELS IN DemoSpectralClustering

This demo implements normalized spectral clustering, using the eigenvectors of the random walk Laplacian L_rw = D^{-1} (D - S). The eigenvectors are used to map each data point i to the new representation (v_1i, v_2i, ..., v_ki), which yields an embedding of the data points into Euclidean space. The final clustering is done using k-means clustering in the new representation. Of course there are many other variants of spectral clustering which we did not incorporate in the demo. For a survey of some of them see the overview paper A tutorial on spectral clustering .