Attacking Evolutionary Algorithms via SparseEA

Over the past few decades, Evolutionary Computation (EC) has been well-developed and widely applied in real-world scenarios, from Logistics Programming, Portfolio Optimization, and Optimal Power Flow to Cancer Radiotherapy Planning. As the society environment becomes even more complex, imperceptible cyber attacks aimed at bottom algorithms are no longer uncommon. However, there is a shortage of attention on adversarial attacks on evolutionary algorithms. For example, a malicious manipulation of datasets can result in a severe performance decline of evolutionary algorithms, significantly impacting the user experience in practical applications. To bridge this gap, this paper introduces a framework for adversarial attacks on evolutionary algorithms aimed at countering both single-objective and multi-objective attacked algorithms of real-world optimization problems. To satisfy the requirement of attack effect and imperceptibility, SparseEA is adopted as the attack algorithm. The experimental results on attacking multiple real-world problems demonstrate the effectiveness of the proposed framework.