<ul><li>Long-range interactions in graph neural networks are challenging and lack robust theoretical foundation.</li><li>Current empirical approaches, such as the Long Range Graph Benchmark, highlight the need for a more principled characterization of long-range problems.</li><li>Researchers have formalized long-range interactions in graph tasks and introduced a range measure for operators on graphs to address this gap.</li><li>This work aims to advance the understanding and evaluation of long-range problems in graph tasks and provide a framework for assessing new datasets and architectures.</li></ul>

On Measuring Long-Range Interactions in Graph Neural Networks

Discover more