Context matters: using reinforcement learning to develop human-readable, state-dependent outbreak response policies.

Area of Science:

• Epidemiology
• Computational Biology
• Artificial Intelligence

Background:

• Real-world infectious disease epidemics are complex with numerous possible scenarios.
• Optimal average control strategies may not be best for every individual epidemic.
• Reinforcement learning (RL) offers a method for developing context-dependent solutions in complex systems.

Purpose of the Study:

• To apply RL for developing context-dependent outbreak response policies.
• To minimize the impact of infectious disease outbreaks, specifically foot-and-mouth disease.
• To translate complex RL policies into human-interpretable heuristics.

Main Methods:

• Utilized reinforcement learning (RL) and Monte Carlo control.
• Developed context-dependent policies for outbreak response.
• Illustrated methods for translating machine-readable policies into simple rules.

Main Results:

• Context-dependent policies adapted to specific outbreaks resulted in smaller disease spread compared to static policies.
• Demonstrated the feasibility of using RL for dynamic epidemic control.
• Showcased two approaches for heuristic translation of RL policies.

Conclusions:

• RL can generate effective, adaptive strategies for infectious disease outbreak control.
• Context-dependent policies derived from RL outperform static approaches.
• Translating RL policies into heuristics aids human decision-making in epidemic response.