III: Medium: A Data-driven and AI-augmented Framework for Collaborative Decision Making to Combat Infectious Disease Outbreaks

新型冠状病毒病（COVID-19）大流行等传染病暴发涉及随时间和空间演变的局部条件，这对政策和决策者来说是一项艰巨的任务，他们需要在不同规模上寻找最佳的非药物干预（NPI）策略，以平衡流行病学效益和社会经济成本。为了帮助解决这一具有挑战性的问题，通过利用数据革命和提高人工智能（AI）的能力，这一多学科项目旨在设计和开发一个数据驱动和人工智能增强的框架，该框架根据不断变化的当地条件量身定制，使自适应国家机构的专家能够有效应对流行病的动态，同时平衡多方面的社会经济影响。拟议的工作不仅有利于地方和联邦政府、地区社区、企业、社会领袖和公众，有助于有效应对公共卫生问题，同时减轻负面的社会经济影响和各种引发的危机，而且还将促进基于科学的强大决策支持系统的发展，以应对未来的自然或人为灾害。这项研究将有利于多学科领域，包括数据科学、机器学习、流行病学、经济学、社会和行为科学。结果（例如，开源代码、数据和模型）将通过出版物、媒体出版社等公开访问并广泛分发。该项目将把研究与教育结合起来，包括新课程开发、学生指导、专业培训和劳动力发展，以及针对代表性不足群体的K-12外展活动。为了通过强有力的应对规划来应对传染病暴发，该项目包括四个相互关联的研究组成部分，以开发一个智能和互动的决策支持框架，允许在潜在的实地实施阶段之前对广泛可能的国家主动行动方案进行计算机探索。首先，该团队将开发一种新的时空异构图模型来抽象利用多源数据的动态。其次，该团队将开发新的技术，通过整合空间和时间依赖关系，同时保留异质性，来学习构建图上的节点（即区域）表示。第三，基于学习到的节点表示，给定一组npi，团队将设计和开发一种创新的npi感知多头变压器，用于多任务预测（即预测流行病动态和相关的社会经济影响）。第四，基于预测，该团队将开发一种具有逆奖励学习的新型多智能体强化学习模型，使专家在循环中找到最优的顺序npi，在政策和决策者设定的某些约束和目标下平衡流行病学效益和社会经济成本。该研究将通过开发一系列原创作品，包括在异构和动态图结构背景下的新型深度图学习技术，推动信息集成和信息学领域的发展，这也将为解决未来自然或人为灾害的类似挑战提供基础工作。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。

Infectious disease outbreaks, such as the novel coronavirus disease (COVID-19) pandemic, entailed localized conditions with evolution in time and space present a daunting task for policy and decision makers in finding optimal non-pharmaceutical intervention (NPI) strategies at different scales that balance epidemiological benefits and socioeconomic costs. To help tackle this challenging problem, by harnessing the data revolution and advancing capabilities of artificial intelligence (AI), this multidisciplinary project aims to design and develop a data-driven and AI-augmented framework that is tailored to the evolving localized conditions and enables expert-in-the-loop for adaptive NPIs to effectively respond to the dynamics of epidemic while balancing the multidimensional socioeconomic impacts. The proposed work will not only benefit local and federal governments, regional communities, corporations, societal leaders and the public by assisting with effective responses to the public health issues while mitigating negative socioeconomic impacts and various induced crises, but will also facilitate the development of robust science-based decision support systems responding to future natural or man-made disasters. The research will be beneficial to multidisciplinary areas, including data science, machine learning, epidemiology, economics, social and behavioral sciences. The outcomes (e.g., open-source code, data, and models) will be made publicly accessible and broadly distributed through publications, media presses, etc. This project will integrate research with education, including novel curriculum development, student mentoring, professional training and workforce development, and K-12 outreach activities aimed at underrepresented groups.To combat infectious disease outbreaks with robust response planning, this project includes four interconnected research components to develop an intelligent and interactive decision support framework that allows in silico exploration of extensive possible NPIs prior to the potential field implementation phase. First, the team will develop a novel spatial-temporal heterogeneous graph model to abstract dynamics of harnessed multi-source data. Second, the team will develop new techniques to learn node (i.e., area) representations over the constructed graph by integrating both spatial and temporal dependencies while preserving the heterogeneity. Third, based on the learned node representations, given a set of NPIs, the team will design and develop an innovative NPI-aware multi-head transformer for multi-task prediction (i.e., forecasting epidemic dynamics and associated socioeconomic impacts). Fourth, based on the predictions, the team will develop a novel multi-agent reinforcement learning model with inverse reward learning to enable expert-in-the-loop in finding optimal sequential NPIs that balance epidemiological benefits and socioeconomic costs under certain constraints and objectives set by policy and decision makers. The research will advance the field of information integration and informatics through the development of a series of original works including novel deep graph learning techniques with the context of heterogeneous and dynamic graph structures, which will also provide foundational work for addressing similar challenges for future natural or man-made disasters.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.