When deep learning meets complex network robustness assessment

doi:10.1016/j.chaos.2026.118837

When deep learning meets complex network robustness assessment

2026-07-24

0

PRE

AI

DOI：10.1016/j.chaos.2026.118837

Original

Original request for help

Save

Abstract

En 中文

Robustness assessment of complex networks is a fundamental issue in network science, with wide-ranging implications for infrastructure resilience, communication systems, and social dynamics. Traditional approaches, however, face critical limitations. Simulation-based methods require repeated execution of computationally intensive procedures, which hinders their ability to provide real-time robustness evaluation. Analytical methods based on percolation theory, while theoretically rigorous, demand complicated mathematical derivations that become infeasible for highly intricate or multilayer networks. To address these challenges, this paper proposes a novel data-driven framework for real-time network robustness assessment. The core idea is to leverage existing robustness evaluation techniques to construct a labeled training dataset. A convolutional neural network (CNN) is then trained on this dataset, enabling the trained model to directly predict robustness values for unseen networks without the need for costly simulations or complex analytical derivations. To validate the proposed method, extensive experiments are conducted on both synthetic and real-world networks. The results demonstrate that the proposed approach achieves accurate predictions of network robustness while significantly reducing computation time.This work encodes both network topology and node perturbation information into a CNN,thereby opening new directions for efficient and scalable robustness analysis in complex networks.

When deep learning meets complex network robustness assessment

Abstract

Journal

Organization