NRPP: A Learning Graph Representation Approach for Network Robustness Prediction

In the field of modern network science, robustness is a key factor in evaluating the characteristics of complex networks. Connectivity robustness and controllability robustness are two important measures. They refer to a network's ability to maintain connectivity and controllability during malicious attacks or random failures. Traditional methods for assessing network robustness, which typically involve time-consuming attack simulations, often suffer from limited accuracy and computational inefficiency. Thus, this paper proposes a simple yet effective method, NRPP, for predicting network robustness using a learning graph representation. This method transforms local nodal information into a graph representation and extracts multi-scale features. Extensive experiments on undirected synthetic networks show: 1) NRPP effectively combines node sorting (NR) with pyramid pooling (PP) to obtain graph-level vector representations, improving network robustness predictions. Ablation experiments validate the necessity of node sorting and pyramid pooling. 2) Experimental results demonstrate that NRPP outperforms three state-of-the-art CNN-based models in predicting network robustness.