Graph Learning: Principles, Challenges, and Open Directions

Official tutorial at ICML 2024, The 41st International Conference on Machine Learning.

Graph Machine Learning has rapidly evolved from early spectral and embedding methods to highly expressive deep message-passing architectures. This tutorial provides a principled and research-grounded overview of the field, designed to connect theoretical foundations with modern empirical practice.

We revisit the core principles underpinning graph learning, including graph signal processing, message passing, and representation learning, and examine how these paradigms shape contemporary Graph Neural Network (GNN) design. Particular attention is devoted to structural and information-theoretic limitations such as oversmoothing, oversquashing, heterophily, and long-range dependency modeling, highlighting both established findings and open debates.

Speakers

• Adrián Arnaiz-Rodríguez, ELLIS Alicante
• Ameya Velingker, Google Research

Panel Discussion (Round Table): Future Directions of Graph Learning

The tutorial concluded with a panel discussion on future directions in graph machine learning, including current limitations, graph foundation models, and opportunities for integrating graph learning with large language models (LLMs).

Panelists:

• Michael Bronstein, DeepMind Professor of Artificial Intelligence, University of Oxford.
• Michael Galkin (online), AI Research Scientist, Intel Labs.
• Christopher Morris, Assistant Professor, RWTH Aachen University.
• Bryan Perozzi, Research Scientist, Google Research.

Moderated by: Adrián Arnaiz-Rodríguez and Ameya Velingker.

Tutorial Goals

The tutorial aimed to equip researchers and practitioners with:

• A consolidated view of the graph learning landscape.
• Conceptual tools to reason about architectural trade-offs.
• A synthesis of theoretical limitations and mitigation strategies.
• A roadmap of open problems and emerging research directions.

Resources and Materials

• ICML site: https://icml.cc/virtual/2024/tutorial/35233
• Tutorial website: https://icml2024graphs.ameyavelingker.com/
• ELLIS Alicante dissemination page: https://ellisalicante.org/tutorials/GraphLearningICML2024/
• Youtube recording of the tutorial: https://www.youtube.com/watch?v=Rd_8QcPg6kw