Efficient algorithms for statistical inference on graphs: from social networks to natural hazard management
Many complex systems from fields as diverse as social networks, mathematical biology, or smart electrical power grids can be modeled as a graph of statistical correlations. The task of making a decision based on such a model (e.g., identifying an imminent hazard, or predicting an oncoming surge of activity on social networks) may appear as being hopelessly complicated and out of reach. This presentation describes a few advanced algorithms that greatly reduce the complexity of decision-making, by following the evolution of "beliefs" (or Bayesian data) on the graph.