Community Detection via Fused Loadings Principal Component Analysis

Monday Sandwich Series

Community detection is one of the most widely studies problems in network research. In an undirected graph, communities are regarded as tightly-knit groups of nodes which are loosely connected between themselves. Among recent developments, spectral clustering and variants thereof have proved some of the most successful techniques for community detection. However, the success of spectral clustering relies strongly on the network density, which is not always a satisfactory feature of large-scale real-world datasets.

To tackle this problem with sparse network datasets, we propose a modification of the spectral clustering method, called Fused Loadings Principal Component Analysis (FLPCA). Following a brief introduction to community detection, I will present the FLPCA algorithm, as well as numerical and some preliminary theoretical results.

This is joint work with Dr Yang Feng (Columbia University) and Dr Richard Samworth 

(University of Cambridge).


Monday, September 16, 2013 - 12:30
to 13:30