【Master Forum】Riemannian Geometry for Next-Generation Graph Foundation Models

May 12, 2026 Upcoming

Topic:?Beyond Words: Riemannian Geometry for Next-Generation Graph Foundation Models

Speaker: Professor Shilun Yu

Host: Professor Xuemin Lin

Date: Tuesday, May 12, 2026

Time: 4:15 PM–5:15 PM

Venue: SIN Wai Kin International Conference Centre (W201), Administration Building

Language: English

Abstract:?

Graph Foundation Models (GFMs) are poised to transform graph learning, yet substantial debate remains over how to construct a general-purpose GFM analogous to Large Language Models (LLMs). Traditional Graph Neural Networks (GNNs) struggle with memory retention, principled interpretability, and multi-domain adaptation, while graph serialization limits the direct application of LLMs, as linear tokens fail to capture the rich structural complexity of graphs. In contrast, Riemannian geometry provides a natural, mathematically principled framework for modeling graph structures, offering compatibility with semantic graph learning and LLM integration. In this talk, we argue that for graphs, geometry speaks louder than words. We introduce Riemannian Graph Foundations, which prioritize intrinsic graph geometry and endow models with endogenous capabilities for structural inference and generation, moving beyond simple representation-space transformations. Preliminary results on building a graph foundation model within Riemannian geometric space will be presented, highlighting how this approach lays the groundwork for next-generation graph intelligence.

Speaker Profile:

Prof. Shilun YU is a Fellow of the ACM, IEEE and AAAS. He received the ACM SIGKDD 2016 Innovation Award for his influential research and scientific contributions to mining, fusion, and anonymization of big data; the IEEE Computer Society 2013 Technical Achievement Award for “pioneering and fundamentally innovative contributions to the scalable indexing, querying, searching, mining, and anonymization of big data”; and the Research Contributions Award from ICDM in 2003 for his pioneering work in data mining. He has published over 1,700 refereed conference and journal papers, which have been cited more than 255,000 times, with an H-index of 216. He has also applied for more than 300 patents. He served as Editor-in-Chief of ACM TKDD (2011-2017) and IEEE TKDE (2001-2004).

?