Learning to Boost Resilience of Complex Networks via Neural Edge Rewiring

Abstract

The resilience of complex networks refers to their ability to maintain functionality in the face of structural attacks. This ability can be improved by performing minimal modifications to the network structure via degree-preserving edge rewiring-based methods. Existing learning-free edge rewiring methods, although effective, are limited in their ability to generalize to different graphs. Such a limitation cannot be trivially addressed by existing graph neural networks (GNNs)-based learning approaches since there is no rich initial node features for GNNs to learn meaningful representations. In this work, inspired by persistent homology, we specifically design a variant of GNN called FireGNN to learn meaningful node representations solely from graph structures. We then develop an end-to-end inductive method called ResiNet, which aims to discover resilient network topologies while balancing network utility. ResiNet reformulates the optimization of network resilience as a Markov decision process equipped with edge rewiring action space. It learns to sequentially select the appropriate edges to rewire for maximizing resilience. Extensive experiments demonstrate that ResiNet outperforms existing approaches and achieves near-optimal resilience gains on various graphs while balancing network utility.