Mitigating epidemic spread in complex networks based on deep reinforcement learning

Area of Science:

• Network Science
• Epidemiology
• Computational Science

Background:

• Complex networks are vulnerable to contagious cascades, necessitating efficient epidemic mitigation.
• Physical quarantine is effective but can incur significant economic costs if not strategically implemented.

Purpose of the Study:

• To develop an innovative, cost-effective strategy for selecting quarantine targets in complex networks.
• To minimize both epidemic spread and quarantine-related economic repercussions.

Main Methods:

• Modeling epidemic spread using Markov chains with stochastic transitions and node quarantines.
• Employing deep reinforcement learning (DRL), specifically the proximal policy optimization algorithm, to train a quarantine strategy.
• Conducting simulations on synthetic and real-world network datasets.

Main Results:

• The DRL-based quarantine strategy effectively controls epidemic spread in complex networks.
• A non-linear relationship was observed between the daily maximum quarantine scale and mitigation effect, showing diminishing returns after a critical threshold.
• The strategy successfully balances infection rate reduction with quarantine costs.

Conclusions:

• Deep reinforcement learning offers a powerful approach for optimizing epidemic mitigation strategies in complex networks.
• Understanding the non-linear impact of quarantine scale is vital for efficient resource allocation and policy-making.
• The proposed method provides a framework for economically viable epidemic response planning.