RAPID: Collaborative Research: Modeling and Learning-based Design of Social Distancing Policies for COVID-19

人类接触是包括COVID-19在内的任何传染病传播的基础。针对COVID-19，广泛实施的社会距离政策正是为了大幅减少个人旅行和由此产生的接触。在一些国家，这些政策有效地减少了感染高峰人数。这些政策也给社会、经济和人民生活带来了巨大的代价：美国经济基本陷入停滞，失业人数已超过2008-2009年金融危机最严重时期。这个快速的COVID-19应用程序将开发一个新的元人群水平模型来模拟COVID-19的传播，并利用强化学习来探索社交距离的最佳聚集限制政策。技术方法将开发SIQR（易感、感染、隔离和恢复）模型，并与强化学习相结合，用于持续监测和政策调整。SIQR模型建立在SIR（易感、感染和恢复）和SEIR（易感、暴露、感染和恢复）模型的经典文献基础上，增强了它们捕捉COVID-19独特隔离特征的能力。提出的项目的重点是SIQR模型与强化学习的连接，以实现一个控制回路，该控制回路在稀疏和噪声观测的情况下提供最优策略。这是对这一新兴的传染病建模和控制的跨学科科学的重要贡献。该项目的成果将对设计COVID-19应对措施具有直接重要性，并有助于更广泛地发展涉及传染病建模、强化学习和大数据机器学习的跨学科教育和研究项目。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Human contacts underlie the spread of any infectious diseases including COVID-19. For COVID-19, the widely implemented social distancing policies are designed precisely to drastically reduce individual travels and the resulting contacts. In a number of States, these policies have effectively reduced the peak number of infections. These policies have also come with huge costs on the society, economy and people’s lives: US economy has largely come to a halt and the number of unemployment claims has now exceeded the worst of the 2008-2009 financial crisis. This rapid COVID-19 application will develop a novel meta-population level model simulating the spread of COVID-19 and utilize reinforcement learning to explore optimal congregation restriction policies for social distancing. The technical approach will develop an SIQR (Susceptible, Infected, Quarantined, and Recovered) model integrated with reinforcement learning for continuous monitoring and policy adjustment. The SIQR model is built on the classic literature of the SIR (susceptible, infectious and recovered) and SEIR (susceptible, exposed, infectious, and recovered) models and enhances their capability to capture the unique quarantine features for COVID-19. The key focus of the proposed project is on the connection of the SIQR model to reinforcement learning to realize a control loop that provides optimal policy in spite of sparse and noisy observations. This is an important contribution to this emerging, interdisciplinary science of infectious disease modeling and control. The results of this project will have both immediate importance for designing the response to COVID-19 and also contribute to the broader development of an interdisciplinary education and research program involving infectious disease modeling, reinforcement learning and machine learning of big data.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.