About this presentation
In this talk, I will describe how to obtain more precise Monte Carlo estimates of expectations using control variates. I will describe control variates in a broad context before delving into some broadly applicable control variates that make use of gradient information. I will focus on two novel control variate approaches: regularised zero-variance control variates (regularised ZV-CV) and semi-exact control functionals (SECF). SECF is particularly beneficial in the context of biased sampling and regularised ZV-CV is competitive in high dimensions. I will use multiple Bayesian inference examples to illustrate how novel and existing control variates can give massive reductions in mean square error.
About the presenter
Dr Leah South is a lecturer in statistics at QUT, an associate investigator of the CDS and an associate investigator of ACEMS. Leah’s research interests are in Bayesian computational statistics, especially variance reduction techniques, scalable Monte Carlo and approximate Bayesian computation. More information about Leah’s research can be found at her website https://sites.google.com/view/leahsouth.
Details:
| Location: | Online |
| Start Date: | 29/04/2022 [add to calendar] |
| Start Time: | 2pm |
| End Date: | 29/04/2022 |
| End Time: | 3pm (AEST) |
| Register: |