Скачать или смотреть Graph Neural Network, Geometric Flows, and Neural DiffusionEquations

Graph Neural Networks (GNNs) have recently become a standard
tool in the machine learning instrumentarium, with applications ranging
from social science to particle physics and drug design. Traditionally,
GNNs have been built on graph theoretical tools such as the
Weisfeiler-Lehman isomorphism tests. In this talk, I will make connections
between GNNs and non-Euclidean diffusion equations. I will show that
drawing on methods from the domain of differential geometry and algebraic
topology, it is possible to describe the expressive power of GNNs, provide
a principled view on such architectural choices as positional encoding and
graph rewiring, deal with heterophilic datasets, as well as explain and
remedy the phenomena of oversmoothing, oversquashing, and bottlenecks.

Michael Bronstein is the DeepMind Professor of AI at the University of
Oxford and Head of Graph Learning Research at Twitter. He was previously a
professor at Imperial College London and held visiting appointments at
Stanford, MIT, and Harvard, and has also been affiliated with three
Institutes for Advanced Study (at TUM as a Rudolf Diesel Fellow
(2017-2019), at Harvard as a Radcliffe fellow (2017-2018), and at Princeton
as a short-time scholar (2020)). He is the recipient of the Royal Society
Wolfson Research Merit Award, Royal Academy of Engineering Silver Medal,
five ERC grants, two Google Faculty Research Awards, and two Amazon AWS ML
Research Awards. He is a Member of the Academia Europaea, Fellow of IEEE,
IAPR, BCS, and ELLIS, ACM Distinguished Speaker, and World Economic Forum
Young Scientist. In addition to his academic career, Michael is a serial
entrepreneur and founder of multiple startup companies, including Novafora,
Invision (acquired by Intel in 2012), Videocites, and Fabula AI (acquired
by Twitter in 2019).