Song, Zexi. Advanced statistical computing: Hamiltonian assisted Markov chain Monte Carlo and imputation maximization stochastic approximation. Retrieved from https://doi.org/doi:10.7282/t3-4zea-3d22
DescriptionIn this dissertation we develop novel methods in two areas of advanced statistical computing. The first area is Markov chain Monte Carlo (MCMC) and the second one is stochastic approximation (SA).
For the MCMC part, we develop and investigate in depth over two chapters, a broad class of irreversible sampling algorithms, called Hamiltonian assisted Metropolis sampling (HAMS). Furthermore, we formulate a framework of generalized Metropolis-Hastings sampling, which not only highlights our construction of HAMS at an abstract level, but also facilitates possible further development of irreversible MCMC algorithms. Extensive numerical experiments are conducted where the proposed HAMS algorithms consistently outperform existing methods.
For the SA part, we propose a new method called imputation maximization stochastic approximation (IMSA) which is suitable for estimation in latent variable models. We focus on its application in generalized linear mixed models and demonstrate its advantages over existing likelihood based methods through simulation studies.