Tag
This paper proposes PairAlign, a pair-centric graph rewiring framework that uses optimal transport-guided communication alignment to alleviate over-squashing in message-passing neural networks, with theoretical analysis and experiments on standard benchmarks.