School of Mathematical Sciences Colloquium Series |
Probability density estimation by
diffusion
by
Prof Dirk Kroese
Location: 7.15 Innova21
Date: Friday, 10 June
Time: 15:10
Abstract: One of the beautiful aspects of Mathematics is that seemingly disparate areas can often have deep connections. This talk is about the fundamental connection between probability density estimation, diffusion processes, and partial differential equations. Specifically, we show how to obtain efficient probability density estimators by solving partial differential equations related to diffusion processes. This new perspective leads, in combination with Fast Fourier techniques, to very fast and accurate algorithms for density estimation. Moreover, the diffusion formulation unifies most of the existing adaptive smoothing algorithms and provides a natural solution to the boundary bias of classical kernel density estimators. This talk covers topics in Statistics, Probability, Applied Mathematics, and Numerical Mathematics, with a surprise appearance of the theta function. This is joint work with Zdravko Botev and Joe Grotowski.
The Colloquium will be followed by a reception for our speaker in
the Staff Tea Room with wine and nibbles to which all are invited.