A New Way to Train AI on Graph Data Without Supervision

Researchers introduce Graphical Mutual Information (GMI), a new way for AI to learn from graph data without labels. By maximizing information between node features, structure, and embeddings, it captures richer patterns than prior methods like DGI. The result: better performance on tasks like node classification and link prediction—even rivaling supervised models.