Revisiting Pre-Propagation GNNs: Robust Diffusion Operators and Hidden-State Re-Propagation

2026-05-24

AI Summary

This work addresses the limited expressiveness and subpar performance of pre-propagation graph neural networks (PPGNNs) on heterophilic graphs, which hinders their practical applicability. To overcome these limitations, the authors propose a novel robust graph diffusion operator for preprocessing, coupled with a few-shot hidden state re-propagation mechanism that enhances model expressiveness during training. By integrating dense node transformations with an efficient mini-batch training strategy, the proposed method substantially narrows the performance gap between PPGNNs and message-passing GNNs. Extensive experiments demonstrate that the approach achieves competitive or even superior accuracy across multiple benchmark datasets while preserving the inherent training efficiency advantage of PPGNNs.

Abstract

Pre-propagation graph neural networks (PPGNNs) decouple node feature propagation from transformation: graph diffusion is performed once as preprocessing, and training reduces to dense per-node transformations. This design enables mini-batch training without inter-node dependencies, avoids repeated sparse matrix--matrix multiplications, and better matches modern accelerators optimized for dense compute. However, their expressivity remains unclear, and empirical results show a gap between PPGNNs and their message-passing counterparts on commonly used graph benchmarks, especially heterophilic ones. In this paper, we propose a suite of robust graph diffusion operators for preprocessing and a few-shot hidden-state re-propagation scheme during training. Our methods improve the validation and test accuracy of PPGNNs, enabling them to match the accuracy of message-passing GNNs while maintaining training efficiency.

Problem

• pre-propagation GNNs
• expressivity
• heterophilic graphs
• graph diffusion
• message-passing GNNs

Innovation

• pre-propagation GNNs
• robust diffusion operators
• hidden-state re-propagation
• heterophilic graphs
• training efficiency