NAVER LABS Europe seminars are open to the public. This seminar is virtual and requires registration
Date: 15th November 2022, 10:00 am (CEST)
Optimal transport for graph-signal processing and learning
About the speaker: Nicolas Courty is Full Professor at University Bretagne Sud since 2018. He obtained his PhD degree in 2002 from INSA Rennes and his ‘Habilitation à diriger des recherches’ in 2013 on computer graphics and animation (avatars, crowds), with a specialization in data-driven methods. He now leads the Obelix team at IRISA, dedicated to machine learning and its applications to Earth Observation. He is an experienced researcher in the domain of machine learning and AI. Among others, he has published several papers in top tier machine learning conferences (NeurIPS, ICLR, ICML, AISTATS, etc.), computer vision (IEEE TPAMI, ECCV, ACCV) and remote sensing (IEEE TGRS, ISPRS journal). From 2014, he has developed expertise in the domain of optimal transport and related applications to machine learning. He is also one of the recipients of the U.V. Helava Award, awarded by the International Society for Photogrammetry and Remote Sensing (ISPRS), for the best paper in the ISPRS journal in years 2012–2015. Nicolas Courty is a member of the European ELLIS society. From 2020, he pilots an ANR Chair program on AI (OTTOPIA), on the topic of applied optimal transport for Remote Sensing.
Abstract: In this talk I will discuss how a variant of the classical optimal transport problem, known as the Gromov-Wasserstein distance, can help in designing learning or data analysis tasks over graphs. After reviewing theoretical and methodological motivations, I will notably present two recent applications to be published at NeurIPS 2022 on the alignment of inter-individual functional brain images, and a new pooling layer for graph neural networks, which achieves state of the art results on graph classification benchmarks.