Contact Tracing and Epidemic Intervention via Deep Reinforcement Learning

The recent outbreak of COVID-19 poses a serious threat to people’s lives. Epidemic control strategies have also caused damage the economy by cutting off humans’ daily commute. In this article, we develop an Individual-based Reinforcement Learning Control Agent (IDRLECA) search for smart epidemic that can simultaneously minimize infections and cost mobility intervention. IDRLECA first hires infection probability model calculate current each individual. Then, probabilities together with individuals’ health status movement information are fed novel GNN estimate spread virus through human contacts. estimated risks used further support RL agent select individual-level epidemic-control actions. training is guided specially designed reward function considering both intervention effectiveness control. Moreover, design constraint control-action selection eases its difficulty improve exploring efficiency. Extensive experimental results demonstrate suppress at very low level retain more than 95% mobility.