Join our slack and come to the next Graph Learning on Wednesdays (GLOW) session.
https://sites.google.com/view/graph-learning-on-weds
On March 26th, 2025, we had the pleasure to host Petar Veličković on the topic of "LLMs as Graph Neural Networks".
Abstract:
Problems such as efficient (length) generalization are highly relevant for building agentic systems, yet they have puzzled LLM researchers for several years. However, they can sometimes be obvious through the lens of graph representation learning. This is because LLMs are, by their design, vulnerable to generalization issues from a variety of geometric perspectives.
A typical GNN researcher will already likely have a solid understanding of most of these perspectives! In this talk, we will use this knowledge, and attempt to scale the mountain of LLM generalization.