Variational Bayes learning over multiple graphs
M. Shiga, Hiroshi Mamitsuka · 2010
Learning (or mining) patterns in graphs has become an important issue in a lot of applications, including web, text and biology. Our issue is graph clustering, i.e. clustering nodes (examples) in a given network. We deal with a situation that we have multiple graphs, sharing nodes but having different edges, where each graph can have only part of the entire true clusters which we call localized clusters, being found in only part of all given graphs. For this issue, we present a probabilistic generative model and its robust learning scheme, being based on variational Bayes estimation. We empirically demonstrate the effectiveness of the proposed framework by using synthetic and real graphs.