Approximate Inference Algorithms in Large Networked Systems
Chongning Na · mediaTUM – the media and publications repository of the Technical University Munich (Technical University Munich) · 2010
Estimation and optimization in large networked system is not a trivial task due to the large amount of participants, the complex uncertainties and processes. Probabilistic models provide a powerful tool for organizing all sources of information and describing explicitly the statistical structure of the estimation problem. Estimation is performed by inferring the posterior probability distributions of the variables that are related to the quantities of interest, given the observation variables. The focus of this thesis is the development of distributed and approximate inference algorithms that solve the estimation problems in large networked systems. These new methods are based on the existing inference algorithms but extend them to adapt to the specific properties of applications in networked systems. The developed methods are applied to solve two practical problems: node localization and clock synchronization.