Damir Filipović: Kernel methods with applications in finance and statistics

Location

FR4, Albanova

Schedule

14:15–15:00 Pre-colloquium by Christian Emmel  in FB55.

15:15–16:15 Colloquium lecture by Damir Filipović.
16:15–17:00 SMC social get together with refreshments.

Abstract

Reproducing kernel Hilbert spaces (RKHS) provide an elegant framework for non-parametric learning of functions. A RKHS is uniquely determined by its reproducing kernel, which can be any symmetric positive semidefinite function on a given set of covariates. In this talk, I first review some basics about kernel-based learning, including the representer theorem, finite-sample guarantees, and low-rank approximations. I then discuss some recent applications in finance and statistics, including dynamic option pricing, term structure estimation, and estimation of Radon-Nikodym densities.