On Friday October 18, 2024 Prof. Enrique Zuazua will talk on “Data Representation by Neural Networks: A control perspective” at the ICMAM 2024, ICMAM Latin America 2024, a hybrid mode conference promoted by the research organisation ICMAM Latin America organized at several universities in Colombia and Mexico from October 14-18, 2024.
Abstract. In this lecture I shall present some recent work in cooperation with Dr. Kang Liu (FAU). We investigate several non-convex optimization problems involving shallow neural networks (NN), with the aim of representing data sets. The problems are convexified using a “mean-field” technique and the lack of relaxation gap is proved employing classical “representer theorems”. We also derive generalization bounds providing insight into the choice of optimal parameters. We also derive efficient numerical algorithms, and analyze the analogy with classical control problems for parabolic equations.
Invited speakers
Hiraku Nakajima. University of Tokyo. President of the International Mathematical Union
Pierre-Louis Lions. Collège de France
Michael Ruzhansky. Ghent University and Queen Mary University of London
Peter Sarnak. Princeton University, USA.
Avi Wigderson. Institute for Advanced Study, Princeton USA.
Efim Zelmanov. University of California San Diego, USA.
Enrique Zuazua. Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU)
Annalisa Buffa. École Polytechnique Fédérale de Lausanne
Alicia Dickenstein. University of Buenos Aires
Serena Dipierro. University of Western Australia
Isabelle Gallagher. Ecole Normale Supérieure
Irene Fonseca. Carnegie Mellon University
Claudia Garetto. Queen Mary University of London
Andrea Solotar. University of Buenos Aires
David Barilari. University of Padova
Cristina Pignotti. University of L’Aquila
Barbara Kaltenbacher. University of Klagenfurt
Irene Sabadini. Politecnico di Milano, Italy
Lorena Bociu. NC State University
Cristina Urbani. Universitas Mercatorum, Italy
Helene Frankowska. Institut de Mathématiques de Jussieu- Paris
WHEN
Fri. October 18, 2024 at 16:00H (09:00H local time)
WHERE
Hybrid (On-site / Online)
On-site: Universidad Javeriana, Caldas University, Autonomous Metropolitan University, Universidad del Valle, Universidad Pedagógica y Tecnológica de Colombia and Universidad del Atlántico.
Online: Zoom / YouTube Channel (ICMAM)
|| See more at the official page of the event.
