David Ginsbourger (IDIAP)

Modeling and optimizing set functions via RKHS embeddings

Speaker :

David Ginsbourger

Date :

March 3rd 2020

Abstract:

Abstract: We consider the issue of modeling and optimizing set functions,
with a main focus on kernel methods for expensive objective functions taking
finite sets as inputs. Based on recent developments on embeddings of
probability distributions in Reproducing Kernel Hilbert Spaces, we explore
adaptations of Gaussian Process modeling and Bayesian Optimization to the
framework of interest. In particular, combining RKHS embeddings and positive
definite kernels on Hilbert spaces delivers a promising class of kernels, as
illustrated in particular on two test cases from mechanical engineering and
contaminant source localization, respectively. Based on several collaborations
and notably on the paper “Kernels over sets of finite sets using RKHS
embeddings, with application to Bayesian (combinatorial) optimization” with
Poompol Buathong and Tipaluck Krityakierne (AISTATS 2020, to appear).

Bio:

David Ginsbourger is working mainly as a permanent senior researcher at
Idiap Research Institute where he is heading the Uncertainty Quantification
and Optimal Design group. He is also holding since 2018 a titular professorship
at the University of Bern, where he has been employed as Oberassistent (2010-2014)
and then Dozent (since his habilitation in 2014), teaching statistics topics and
also optimization methods. A significant part of his research deals with Gaussian
random field modeling and adaptive design of experiments. Further interests include
design and estimation of kernels and parameters, as well as connections between
spatial statistics and functional analysis. From the real-world application side,
he has been working with a number of colleagues both from engineering as well as
from geosciences. In the last years, he and his team have started collaborations
with climate scientists, and also recently with colleagues from medicine.