Efficient Epidemic Intervention Generation: A Graph Adversarial Attack Perspective

Epidemic outbreaks spread through networked interactions, making graph-based modeling essential for effective intervention strategies. In this study, we propose EpiAdv, a scalable approach that integrates a modified Susceptible-Infected-Recovered (SIR) model with graph adversarial attack techniques to control infection spread on time-varying graphs. EpiAdv dynamically adjusts node states and edge weights using gradient descent, optimizing an objective function to balance infection reduction and intervention. Experimental results show that EpiAdv significantly outperforms baseline methods, such as expert-based and degree-based heuristics, in reducing infection rates while minimizing isolation costs. Furthermore, adaptability tests demonstrate the robustness of EpiAdv across different intervention timing scenarios, providing an effective and efficient framework for managing epidemics in dynamic networks.