Paper accepted at ICLR

The paper “MaxCutPool: differentiable feature-aware Maxcut for pooling in graph neural networks” by Carlo Abate and Filippo Maria Bianchi has been accepted at ICLR 2025!

The paper proposes a novel approach to compute a MaxCut partition in attributed graphs and can be used to implement pooling in Graph Neural Networks. It works particulary well on heterophilic graphs.

The preprint is available on Arxiv and the code on Github.