Speaker |
Takanori Maehara |
|---|---|
Affiliation |
Roku, Inc. |
Date |
Friday, 29 November 2024 |
Time |
12:00-13:00 |
Location |
Maths 706, Gordon Street (25), University College London, London WC1H 0AY |
Link |
https://ucl.zoom.us/j/99748820264 |
Event series |
Jump Trading/ELLIS CSML Seminar Series |
Abstract |
Graph Neural Networks (GNNs) are models processing graph-structured data, making them valuable for practical tasks such as spam detection in web graphs, link prediction in social networks, and chemical analysis in molecules. They also attract attention from theoreticians due to their connections with various fields such as graph theory, differential geometry, and signal processing. An important research topic is the Expressive Power of GNNs, which examines what functions these networks can represent and learn. In this talk, I will give a brief introduction about GNNs and its expressive power. Then, I'll present our recent result revealing the relationship between the GNN architecture and its expressive power in terms of the graph homomorphisms (will appear in NeurIPS'24). |
Biography |
Takanori Maehara is a Senior Software Engineer at Roku. He received his PhD from the University of Tokyo in 2012 and worked in Japanese academia 8 years as a Postdoctoral Researcher at the National Institute of Informatics (2012-2015), an Assistant Professor at Shizuoka University (2015-2016), and a Unit Leader at the RIKEN Center for Advanced Intelligence Project (2016-2020). He then transitioned to industry in the UK, and worked as a Software Engineer at Facebook (2020-2024) before taking on his current role at Roku. |