Toward Universal Graph Representations - Foundations and Frontiers of Graph Foundation Models

Explore the quest for universal representations in graph foundation models through this lecture that examines the challenging goal of creating effective universal embeddings in modern graph machine learning. Dissect the inherent tensions between positional and structural node embeddings in graphs, and discover how to overcome task-specific symmetries and invariances when developing universal representations. Learn about the role of invariances in statistical tests for addressing challenges posed by distinct attribute domains across different graph datasets. Examine novel applications of graph learning to algorithmic reasoning, with particular focus on real-world network optimization problems. Gain insights into the theoretical foundations and practical frontiers of graph foundation models, understanding how these approaches can bridge the gap between graph learning and theoretical computer science applications.