### Abstract:

Markov chain Monte Carlo(MCMC) is a staple tool in statistics used for sampling from a distribution where direct simulation is not possible (or difficult). In this talk we consider  an acceptance-rejection sampler based on a deterministic driver sequence, which is a simple case of a Markov chain quasi-Monte Carlo (MCQMC) method. The deterministic sequence is chosen such that the discrepancy between the empirical proposal distribution and the proposal distribution is small. We use quasi-Monte Carlo (QMC) point sets for this purpose. In a general setting, we prove an upper bound and a lower bound of the discrepancy of samples generated by the deterministic acceptance-rejection sampler. Improved upper bounds are obtained under certain assumptions.  The numerical experiments verify the theoretical conclusions and show convergence rates beyond plain Monte Carlo.

Speaker

Houying Zhu

Research Area
Affiliation

UNSW

Date

Tue, 17/09/2013 - 11:30am to 11:55am

Venue

RC-4082