 |
Center for Nonlinear Studies |
Graphical models, or Markov Random Fields, have been widely deployed in diverse statistical applications; this talk presents recent progress in their application to a new domain: networks. We present three results that show how graphical model algorithms for estimation, learning and sampling provide new algorithms for (1) resource allocation in wireless and sensor networks, (2) faster (i.e. lower delay) communication networks, and (3) network structure learning. We also show how our network settings seem to provide useful "special cases" which permit progress on an important general endeavour: obtaining detailed theoretical results on the performance of graphical model methods.
Host: Misha Chertkov