School of Physics Fall Colloquium Series-Dr. Jonathan Halcrow

Speaker: Jonathan Halcrow (Alphabet/Google)

Host: Predrag Cvitanovic/ Noel Dudeck

Title:  Reasoning about Graphs with Large Language Models

Abstract: Graphs are a powerful tool for representing and analyzing complex relationships in real-world applications. Large Language Models (LLMs) have demonstrated impressive capabilities by advancing state-of-the-art on many language-based benchmarks. Their ability to process and understand natural language open exciting possibilities in various domains. Despite the remarkable progress in automated reasoning with natural text, reasoning on graphs with LLMs remains an understudied problem that has recently gained more attention. This talk builds upon recent advances in expressing reasoning problems through the lens of tasks on graph data. We will provide an in-depth discussion of techniques for representing graphs as inputs to LLMs, as well as some theoretical limits to the ability of transformers to solve various graph tasks.

Bio: Jonathan Halcrow is a Ramblin Wreck from Georgia Tech, completing his undergraduate studies at GT in 2003 and Physics PhD in 2008, studying under Predrag Cvitanovic. His thesis covered the topic of turbulence in Plane Couette flow. More recently, he has been working as a Software Engineer at Google since 2013, as part of the Graph Mining team in Google Research in Atlanta. His research at Google covers the topics of approximate nearest neighbor search, graph neural networks, and improving the abilities of large language models to reason about structured data.