A multi-factor evolutionary algorithm for solving the multi-tasking robust optimization problem on networked systems

Abstract

Realistic networks constantly face complex environments and are vulnerable to external disturbances that can damage their structure and disrupt information flow. These challenges give rise to two critical optimization problems: network robustness optimization and robust influence maximization. These problems hold significant theoretical and practical importance, garnering increased attention in recent research. Models and optimizers have been developed to tackle related typological design and seed determination tasks, which provide rational candidates for practical scenarios. However, existing approaches have limitations. Firstly, while structural disturbances significantly impact network robustness, their effects on information propagation remain understudied. The quest for robust seed determination in the face of topological failures remains unresolved. Secondly, both network robustness optimization and robust influence maximization focus on leveraging structural network information, but the potential synergy between solutions for these problems has been overlooked in current research. There is a pressing need for highly efficient optimizers and versatile candidate solutions capable of withstanding various scenarios comprehensively. In response to these limitations, this study introduces multi-task optimization theory to address the network correlation optimization problem described above. We first systematically analyze the correlation between the two problems, demonstrating a positive relationship through experimental results. Building on this insight, we propose MFEA-Net, a multifactor evolutionary algorithm equipped with problem-specific operators. MFEA-Net concurrently tackles the challenges of network robustness optimization and robust influence maximization, considering multiple optimization scenarios simultaneously. It aims to uncover synergistic insights across different tasks during the optimization process. Our experimental results indicate that MFEA-Net surpasses existing methods in terms of performance. This research bridges a gap in the fields of robust influence maximization and network robust cooperative optimization, offering valuable guidance for addressing real-world network system design and optimization challenges.