Multiple marginalized population Monte Carlo

Population Monte Carlo (PMC) algorithms iterate on a set of samples and weights to approximate a stationary target distribution. Their estimation quality and convergence efficiency rely on many factors including the number of samples and the choice of importance function. The computational complexity of the PMC algorithm becomes increasingly challenging as the numbers of the unknowns increases. In this paper, we propose a marginalized PMC algorithm for high-dimensional problems, where the state space of the system is partitioned into several subspaces of lower dimensions and handled by a set of marginalized PMC estimators. Simulation results show the accuracy and feasibility of the method as well as its improvement with respect to other conventional approaches.