Please join us for an FI Computational Methods and Data Science Journal Club with Wenda Zhou, CDS/Flatiron Faculty Fellow at NYU CDS and CCM. Wenda's talk will take place from 3:00-4:00 PM in the CCN Classroom and will be followed by a reception from 4:00-5:00 PM in the CCN 4th Floor Pantry.
Title: An introduction to higher-order graph neural networks
Abstract: Graph neural networks have proven to be an effective and practical way of bringing the deep learning revolution to relational data. However, recent work has also shown that they may be fundamentally limited in processing certain types of information, for example, they are limited in power by the Weisfeiler-Lehman hierarchy. To overcome these issues, a variety of extensions, often grouped under the "higher-order graph neural network" designation, have been proposed. After introducing standard message passing neural networks, I will give an overview of recent developments in higher-order architectures, and reflect on the future of graph neural networks.