Henry Hoffmann

We enhance Graph Neural Networks by incorporating randomized global embeddings of node features alongside traditional local representations, which mitigates oversquashing, oversmoothing, and expressivity limitations.