Many problems in different scientific domains can be described through statistical models that relate the observed data to some unobserved parameters. The estimation of the hidden parameters is crucial in order to perform several statistical signal processing and machine learning tasks, such as prediction of the evolution of the system, model selection, feature selection, explanatory modeling, etc. In the Bayesian framework, statistical inference is performed based on the relevant posterior distribution. However, in most real-world problems, this posterior distribution is intractable and must be approximated. Monte Carlo algorithms are remarkably flexible and extremely powerful to solve such inference problems. Nevertheless, in complex non-linear and/or high-dimensional systems, standard Monte-Carlo techniques could lead to unsatisfactory results. This manuscript aims to present my research activities on the development of novel sequential Monte Carlo strategies for Bayesian inference in both static and dynamic systems and then to the application of such approaches to solve challenging problems related to localization and tracking of multiple objects.
defended on 06/12/2017