RAPID: Infer and Control Global Spread of Corona-Virus with Graphical Models

图形在物理、生物和社会系统中对统计数据驱动的现象进行建模时无处不在。COVID-19和其他冠状病毒在人、社区、城市、州和国家之间的传播可以用一个图表来描述，我们称之为冠状图。一般来说，我们打算利用的电晕图将能够描述特定的区域，时空分辨率水平，相关的地理，交通和社会互动细节-例如，亚利桑那州图森市在2020年夏天，以家庭为基本单位，并考虑企业及其当前的孤立状态。该项目的目标是建立一种新的方法，该方法允许在给定的时空规范和数据的情况下（机器）学习电晕图和底层电晕图形模型，然后能够进行插值，即对病毒传播进行社会重要的推理预测。RAPID提案的实质是开发和转移技术，从基础计算机科学、应用数学和统计学的最新技术到COVID-19传播的网络建模。在该项目的支持下开发的技术将允许将流行病学输入（例如以流行病学的房室模型表示的输入）与人工智能和数据科学中最先进的方法（例如图形模型和深度学习）相结合。Corona图形模型将灵活处理异质数据源，随着世界从全球自我隔离的“锤子”阶段恢复并过渡到平衡冲突目标的数月“舞蹈”，即保持病毒的繁殖率，同时最大限度地减少经济和社会成本。该项目的技术目标分为三个方面，分别侧重于电晕图形模型的制定，推理和学习。制定推力枢轴建设一类图形模型分析，在其他两个后续推力，通过整合流行病学，交通和其他相关的考虑因素，变量和约束。第二，推理推力努力将联合收割机现有的方法与新的方法和算法相结合，以回答诸如计算观察地理区域（包括电晕图的许多节点）的边际概率以使免疫群体的密度为10%的复杂性是多少的问题。在第三个信任中，将从可用数据（例如在整个感兴趣的时间段内在电晕图的多个节点收集的病毒暴露样本）中学习电晕图和流行病学有意义的因素（例如节点间交互的频率和节点隔离的效率）的选择。在技术层面上，使用图形模型进行建模、学习和推理的项目方法将为将来自流行病学、交通和社会科学的特定位置和时间的数据和信息结合在一起打开大门。该项目还将向更广泛的基础人工智能社区以及对适应新的应用知情的图形模型方法感兴趣的多个其他研究社区传播图形模型的想法、算法、数据和基准。该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估来支持。

Graphs are ubiquitous in modeling statistical data-driven phenomena in physical, biological and societal systems. The spread of COVID-19, and other corona-viruses, between people, communities, cities, states and countries can be described in terms of a graph, which we refer to as the Corona Graph. Generally, the Corona Graphs we aim to utilize will be able to describe a specific region, level of spatio-temporal resolution, related geographical, transportation and social interaction details - for example, the city of Tucson, Arizona, during the summer of 2020, with a house-hold as an elementary unit, and accounting for businesses and their current state of isolation. This project goal is to build novel methodology which allows, given spatio-temporal specification and data, to (machine) learn the Corona Graph and underlying Corona Graphical Model, which will then be capable of interpolating, i.e. making societally important inference predictions about the virus spread. The essence of the RAPID proposal is in developing and transferring technology from the state-of-the-art in the foundational computer science, applied mathematics and statistics to the network modelling of COVID-19 spread. The technology developed under auspices of the project will allow to mix epidemiological inputs, such as these expressed in terms of the compartmental models of the epidemiology, with the state of the art approaches in AI and Data Science, such as Graphical Models and Deep Learning. The Corona Graphical Models will be flexible in dealing with heterogenous data sources, becoming available as the World recovers from the “hammer” stage of global self-isolation and transitions to the multi-month “dance” of balancing the conflicting objectives of keeping the reproduction rate of the virus under control while also minimizing the economic and societal costs. The technical aims of the project are divided into three thrusts, focused on the formulation, inference and learning, respectively, of the Corona Graphical Models. The formulation thrust pivots construction of a class of Graphical Models analyzed in the two other follow up thrusts by integrating epidemiology, transportation and other relevant considerations, variables and constraints. The, second, inference thrust strives to combine existing methodologies with novel methods and algorithms to answer questions such as, what is the complexity of computing marginal probabilities of observing a geographical area, including many nodes of the Corona Graph, to have density of immune population to be 10%? Selection of the Corona Graph and epidemiology-meaningful factors, such as frequency of inter-node interactions and efficiency of a node quarantine, will be learned in the third trust from available data, such as samples of the virus exposure collected at multiple nodes of the Corona Graph throughout a period of interest. On the technical level the project approach to modeling, learning and inference with Graphical Models will open doors to combining under one umbrella location- and time-specific data and information from epidemiology, transportation and social sciences. The project will also result in the dissemination of Graphical Model ideas, algorithms, data and benchmarks to the broader foundational AI community and also to multiple other research communities interested in adapting the novel application-informed Graphical Model methodology.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.