Deep-learning aided consensus problem constrained by network-centrality

Abstract

The deep-learning aided parameter optimization method for the average consensus problem with a complex network has been proposed by Kishida et al., which can significantly accelerate its convergence performance. However, the optimized parameters cannot be applied to the different network topology from the one used in the training. This work proposes a new optimization method constrained by the restriction caused by the network centrality, in which the optimized weighting factors can apply to the different network topology used in the training. Specifically, this work considers five types of network centralities and discusses which centrality is most suitable for the constraint of the proposed method.