Ranking-free attacks to complex networks

Many real-world systems can be seen as complex networks, where nodes denote elements in the systems and links denote relationships between elements. Since most of the real-world systems are vulnerable under disruptions, it is necessary to perform research on network robustness. However, finding the optimal attack on networks is a NP-hard problem, thus many methods based on ranking certain network metrics are proposed, for their low computational time complexity. Although strategies can be obtained quickly according to the ranking methods, the quality of the strategies is unsatisfying, because no single metric can represent all the properties of a network. In this paper, we propose a ranking-free method to attack networks. The main idea of our method is to identify components, which are likely to be in the final network after attack. Our research targets a trade-off between expensive optimal attacks and fast ranking-based heuristics. The evaluations, both on random networks and real-world networks, show that our method behaves better than popular ranking methods.