Workshop on kernel and sampling methods for design and quantization


When & where

The workshop will (hopefully) take place IRL, on November 15, 2021 at Institut Henri Poincaré, Paris.

Room: amphithéâtre Hermite (ground floor).




Presentation of the workshop

The recent period has seen a considerable increase in the use of kernel methods in various fields of applied mathematics and statistics, with a flourishing expansion of applications in many areas of engineering. This includes in particular, but not limited to, function approximation and integration, and covers many topics of interest to the MASCOT-NUM network. In recent years, minimisation of a kernel discrepancy (the maximum mean discrepancy) has been popularised as a general tool for quantifying a probability distribution and constructing space filling designs, and determinantal point processes have emerged as a very promising tool for generating random sets of points with appropriate repulsion properties.

The workshop aims to introduce these modern topics to the MASCOT-NUM community through a series of presentations by key researchers in the field.

Organizers: Luc Pronzato, Julien Bect.


Registration

Registration is free but appreciated.


Agenda


Speakers

Luc Pronzato (CNRS & Univ. Côte d'Azur, France)
Introduction

Pierre Olivier Amblard (CNRS & Univ. Grenoble Alpes, France)
An introduction to determinantal point processes
GP regression in the flat limit

Jean-François Coeurjolly (Univ. Grenoble Alpes, France)
Repulsiveness for integration (not my social program)

Rémi Bardenet (CNRS & Univ. Lille, France)
Interpolation and experimental design with volume sampling

Onur Teymur (Newcastle Univ. & the Alan Turing Institute, UK)
Optimal quantisation of probability measures using maximum mean discrepancy

Marina Riabiz (King's College London & the Alan Turing Institute, UK)
Kernel Stein discrepancy minimization for MCMC thinning, with application to cardiac electrophysiology