Towards Principled Graph Transformers Luis Müller Daniel Kusuma

However, such architectures often fail to deliver solid predictive performance on real-world tasks, limiting their practical impact. In contrast, global attention-based models such as graph transformers demonstrate strong performance in practice.