Robust training of open-set graph neural networks on graphs with in-distribution and out-of-distribution noise

Key Points

• Robust training improves accuracy in challenging conditions, addressing both in-distribution and out-of-distribution noise.
• Performance metrics indicate a significant increase in robustness, achieving above 80% accuracy under noisy conditions.
• Analysis of graph structures reveals the adaptability of open-set graph neural networks to varying noise levels.
• Highlights the potential for enhanced applications in sectors like social networks and fraud detection.