Pierre Borgnat – P. Borgnat (ENS Lyon) : Graph Wavelets and Multiscale Community Mining in networks

Carte non disponible

Date/heure
Date(s) - 23 mai 2014

Catégories Pas de Catégories


Graph Wavelets and Multiscale Community Mining in networks\n\nBy Pierre Borgnat, ENS Lyon\n\nJoint work with N. Tremblay\n\nFor networks, an important issue is the finding of communities, i.e., groups of nodes that are well connected together, and more than with the rest of the network. A signal processing approach is developed for the multiscale detection of communities in networks. This method relies on a carefully engineered wavelet transform on graphs, so as to introduce the notion of scale and to obtain a local view of the graph from each node. This gives a notion of distance between nodes, thereby enabling to cluster nodes according to their community at the scale of analysis. To make the method suitable for the analysis of large graphs, a collection of random vectors is used to estimate the correlation between the nodes. Finally, a notion of partition stability and an empirical statistical test are introduced, allowing us to assess which scales of analysis of the network are relevant. The method is illlustrated on real data of social networks, on models for signal processing on graphs, and on benchmarks of graphs with multiscale communities.[

Pierre Borgnat – P. Borgnat (ENS Lyon) : Graph Wavelets and Multiscale Community Mining in networks

Carte non disponible

Date/heure
Date(s) - 23 mai 2014

Catégories Pas de Catégories


Graph Wavelets and Multiscale Community Mining in networks\n\nBy Pierre Borgnat, ENS Lyon\n\nJoint work with N. Tremblay\n\nFor networks, an important issue is the finding of communities, i.e., groups of nodes that are well connected together, and more than with the rest of the network. A signal processing approach is developed for the multiscale detection of communities in networks. This method relies on a carefully engineered wavelet transform on graphs, so as to introduce the notion of scale and to obtain a local view of the graph from each node. This gives a notion of distance between nodes, thereby enabling to cluster nodes according to their community at the scale of analysis. To make the method suitable for the analysis of large graphs, a collection of random vectors is used to estimate the correlation between the nodes. Finally, a notion of partition stability and an empirical statistical test are introduced, allowing us to assess which scales of analysis of the network are relevant. The method is illlustrated on real data of social networks, on models for signal processing on graphs, and on benchmarks of graphs with multiscale communities.[