Department Seminar by Mike Ludkovski

Please Note: This seminar will be given online.

Multi-population longevity modeling with Gaussian Processes

I will discuss several interrelated projects on the use of multi-output Gaussian Process (GP) models for analyzing longevity jointly across several populations. The spatial covariance framework  underlying GPs offers a flexible way to capture cross-population dependence within an Age-Period setup. In turn GPs generate rigorous uncertainty quantification, non-parametric forecasts, long-term coherence, and a transparent method for information fusion and extrapolation.  In order to handle 5+ populations, we moreover investigate coregionalization ideas for dimension reduction. Our applications include analysis of Human Mortality Database data across EU nations and genders, analysis of cause-of-death datasets across a handful of countries, and analysis of national Covid-19 excess deaths. Illustrations will include 2 RShiny apps posted on the web.