Mitigating epidemic spread in complex networks based on deep reinforcement learning.

The article focuses on developing an efficient and cost-effective epidemic mitigation strategy in complex networks by applying deep reinforcement learning (DRL). It models epidemic spread as a Markov decision process with nodes representing communities in states of susceptible, exposed, infected, recovered, or quarantined, and uses the proximal policy optimization (PPO) algorithm to train an agent that sequentially selects nodes for quarantine to minimize infection rates and quarantine costs. Simulations on synthetic small-world and real-world dolphin social networks demonstrate that the PPO-based quarantine strategy outperforms traditional degree- and betweenness-centrality-based methods by achieving faster epidemic control with lower quarantine costs. Additionally, the study identifies a nonlinear relationship between the daily maximum quarantine scale and mitigation effectiveness, revealing a critical threshold beyond which increasing quarantine scale yields diminishing returns. This insight aids in optimizing quarantine parameters for practical epidemic control in complex networked communities.