-GNN: A Robust Ensemble Approach Against Graph Structure Perturbation
Graph Neural Networks (GNNs) are playing an increasingly important role in the efficient operation and security of com puting systems, with applications in workload scheduling, anomaly detection, and resource management. However, their vulnerability to network perturbations poses a signif icant challenge. We propose š¯›½-GNN, a model enhancing GNNrobustness without sacrificing clean data performance. š¯›½-GNN uses a weighted ensemble, combining any GNN with a multi-layer perceptron. A learned dynamic weight, š¯›½, modulates the GNNā€™s contribution. This š¯›½ not only weights GNNinfluence but also indicates data perturbation levels, enabling proactive mitigation. Experimental results on di verse datasets show š¯›½-GNNā€™s superior adversarial accuracy and attack severity quantification. Crucially, š¯›½-GNN avoids perturbation assumptions, preserving clean data structure and performance.
