Using Machine Learning for Determining Network Robustness of Multi-Agent Systems Under Attacks

摘要

Network robustness has been the key metric in the analysis of secure distributed consensus algorithms for multi-agent systems (MASs). However, it is proved that determining the network robustness of a MASs with large nodes is NP-hard. In this paper, we try to apply machine learning method to determine the robustness of MASs. We use neural network (NN) that consists of Multilayer Perceptions (MLPs) to learn the representation of multi-agent networks and use softmax as our classifiers. We compare our method with a traditional CNN-based approach on a graph-structured dataset. It is shown that with the help of machine learning method, determining robustness can be possible for MASs with large nodes.