Ralts: Robust Aggregation for Enhancing Graph Neural Network Resilience on Bit-flip Errors

Graph neural networks (GNNs) are increasingly deployed in safety-critical domains such as finance and healthcare, where erroneous predictions can have severe consequences. While prior studies on GNN robustness have primarily focused on software-level threats, hardware-induced faults, particularly transient bit flips and silent data corruption, remain underexplored. As hardware systems scale to advanced technology nodes to meet performance and energy demands, their vulnerability to such faults continues to grow, as reported by major technology companies including Google and Meta. In response, we propose Ralts, a generalizable and lightweight solution to bolster GNN resilience to bit-flip errors. Ralts exploits graph similarity metrics to filter out outliers and recover compromised graph topology, and incorporates these protective techniques directly into aggregation functions to support any message-passing GNNs. Evaluation results show that Ralts consistently improves robustness across diverse GNN models, datasets, error patterns, and both dense and sparse architectures. Ralts achieves execution performance comparable to native PyTorch Geometric aggregation operators. Ralts is available at: https://github.com/ORCA-lab/Ralts.