Caltech Home > PMA Home > Calendar > CMX Lunch Seminar
open search form
Wednesday, January 27, 2021
12:00 PM - 1:00 PM
Online Event

CMX Lunch Seminar

Modeling and optimizing set functions via RKHS embeddings
David Ginsbourger, Titular Professor, Departement of Mathematics and Statistics, University of Bern,
Speaker's Bio:
I am working at the Institute of Mathematical Statistics and Actuarial Science (Department of Mathematics and Statistics, University of Bern, Switzerland) where I am leading a research group and have been teaching a variety of courses. I am a member of the Oeschger Center for Climate Change Research of the University of Bern, on the scientific committees of the GDR Mascot-num and the ANR project RISCOPE, serving as a scientific adviser to the OQUAIDO chair in applied mathematics, and on the editorial board of the SIAM/ASA Journal on Uncertainty Quantification as an associate editor. I defended my venia docendi (habilitation) in Statistics and Applied Probability before the Faculty of Science of the University of Bern in 2014 and my PhD in Applied Mathematics at the Ecole des Mines de Saint-Etienne in 2009. Previous to that, I obtained a double graduate diploma from Ecole des Mines de Saint-Etienne and Berlin Technical University (2005), a research master's degree in Applied Mathematics jointly awarded by Jean Monnet University and Ecole des Mines de Saint-Etienne (2005), and a licence in Mathematics from Joseph Fourier University, Grenoble (2002). From 2008/2009, I worked as an assistant and a scientific collaborator, respectively in the Institute of Mathematics and the Centre for Hydrogeology and Geothermics (Stochastic Hydrogeology Group), University of Neuchatel, before starting to work as Senior Assistant (2010) and then Dozent (2014) at the Institute of Mathematical Statistics and Actuarial Science of the University of Bern. From September 2015 to May 2020, I worked mainly as a permanent senior researcher at Idiap Research Institute where I headed the Uncertainty Quantification and Optimal Design group. During the last two academic years, I lectured at Ecole Polytechnique Federale de Lausanne and participated to the Master in Artificial Intelligence offered by Unidistance and Idiap. I have also held since 2018 a titular professorship at the University of Bern, where I have been employed in several settings since my habilitation. A significant part of my research deals with Gaussian random field modelling and adaptive design of experiments, with a focus on bayesian global optimization and related topics such as Bayesian set estimation. Further interests include design and estimation of covariance kernels and parameters, as well as connections between Kriging and functional analysis approaches (notably the theory or Reproducing Kernel Hilbert Spaces). From the real-world application side, I have been working with a number of colleagues both from engineering and from geosciences. In recent years, my team and I have started collaborations with climate scientists, and now also with colleagues from medecine.

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).

For more information, please contact Jolene Brink by phone at 6263952813 or by email at [email protected] or visit CMX Website.