Home > Events > 2016 Seminars & Colloquia > Michael Lavine, University of Massachusetts

Michael Lavine, University of Massachusetts

Main Content

WHIM: Function Approximation Where It Matters
21 January 2016 from 4:00 PM to 5:00 PM
201 Thomas Building
Contact Name
Contact Phone
Add event to calendar

If f is either a likelihood or posterior density function, it is often of interest to find the region of the parameter space where f is large relative to its maximum.  Typical tools for working with f, such as optimizers and MCMC, can fail, especially when f is multimodal or has a large plateau.  A new algorithm called WHIM — for function approximation WHere It Matters — is guaranteed not to fail for f's arising from a large class of statistical models, even though some of those f's are multimodal or have plateaus.

 This talk will introduce WHIM, illustrate it with one or two examples, and describe what makes it work so we can understand whether it can be applied to other statistical models.  Topics for further research — grad students take note — will be mentioned along the way.


Filed under: ,

Navigation for this Section