Poster
in
Workshop: Machine Learning Multiscale Processes
On the successful Incorporation of Scale into Graph Neural Networks
Christian Koke · Yuesong Shen · Abhishek Saroha · Marvin Eisenberger · Bastian Rieck · Michael Bronstein · Daniel Cremers
Keywords: [ Graph Neural Networks ] [ (Resolution-)Scale ] [ Generalization ]
Standard graph neural networks assign vastly different latent embeddings to graphs describing the same physical system at different resolution scales. This precludes consistency in applications and prevents generalization between scales as would fundamentally be needed in many scientific applications. We uncover the underlying obstruction, investigate its origin and show how to overcome it.