An adaptive attack model to network controllability

For the ultimate goal of protecting the network controllability and enhancing the controllability robustness, one can learn from how a network can be effectively destructed. In this paper, based on Thompson Sampling, we propose a Probabilistic Adaptive Attack model (PAA) that aims at destructing the network controllability. A set of multiple established attack strategies are employed in PAA, and a probabilistic model is used to adaptively choose the most destructive attack strategy for the current attack. The probabilistic model is updated based on the feedback of the controllability destruction of each single attack strategy. After applying PAA on both synthetic and real-world networks, extensive simulation results not only verify that PAA is more destructive than the single attack strategies, but also show that the probabilistic model can effectively choose the true best-performing attack strategy. Its accuracy reaches 81.96% in synthetic networks and 94.14% in real-world networks. The presented study aims to propose a highly efficient attack scheme and encourage the research of advanced methods to build robust networked systems.