Title: Bayesian Kalman filter and Markov Chain Monte Carlo
Speaker: Ye Wang is a PhD candidate of Statistical Science at Duke University. Prior to that he obtained an MSEM degree from Duke and a B.S. in computational mathematics from Harbin Institute of Technology. His research, conducted under the supervision of professor David Dunson, focuses on scalable Bayesian inference and nonparametric models.
Abstract: This talk will cover two important technologies in Bayesian computation. I will first reconstruct Kalman filter via a Bayesian framework called the dynamic linear model (DLM). Several variants of the Kalman filter and their Bayesian interpretations will be presented. I will demonstrate the Bayesian Kalman filter via a case study of the mouse brain network. In the second part of this talk I will cover the arguably most important class of computation techniques in Bayesian world, the Markov chain Monte Carlo (MCMC). I will briefly talk about Metropolis-Hasting, Gibbs sampler, Hamiltonian Monte Carlo and data augmentation. Lastly I will demonstrate the potential mixing issue of MCMC via a probit regression example and present a solution termed the calibrated data augmentation.
Acknowledge: The materials of this talk are partially from the course material of prof. Mike West and prof. Alan Gelfend. The calibrated data augmentation technique is a working project of Leo Duan (post-doc) and prof. David Dunson.
David Dunson: Arts and Sciences Distinguished Professor, Departments of Statistical Science, Mathematics, and ECE, Duke. Recipient of COPSS Presidents’ Award 2010.
Mike West: The Arts & Sciences Professor of Statistics & Decision Sciences, Department of Statistical Science, Duke.
Alan Gelfend: J.B. Duke Professor of Statistics and Decision Sciences, Department of Statistical Science, Duke.