RAPID: SaTC: FACT: Federated Analytics based Contact Tracing for COVID-19

新冠肺炎是一种由新发现的冠状病毒引起的高传染性疾病，其传播已在全球达到大流行水平。随着美国感染、重症监护干预和死亡人数的持续上升，支持接触者追踪(CT)的移动应用程序(APP)正被迅速部署来监控新冠肺炎的传播。然而，基于匿名共享令牌而不利用丰富的本地设备数据的持续部署，在及时监控疾病传播方面是不够的，而且还容易受到隐私和安全攻击。迫切需要开发既能监控又能干预的CT应用程序，在尊重用户安全和隐私的同时限制新冠肺炎的传播。该项目通过基于联合分析的联系人跟踪(FACT)解决这一挑战，这是一种改进的联合学习方法，可以以私有和安全的方式利用设备级数据和服务器功能。FACT通过包括热点识别、用户警报和对用户新冠肺炎风险的持续评估来实现预防和干预。FACT保证了一种私密和安全的方式来(I)评估用户对检测的需求或他们对暴露的恢复能力，以及(Ii)评估整个人群的群体免疫力。FACT提出了一种安全的GPS+蓝牙系统，使服务器能够以保护隐私的方式检测地理感染集群，从而解决了当前基于蓝牙的系统对各种攻击的脆弱性和服务器学习的局限性。FACT还利用丰富的设备级移动性和声音传感数据，使用联合学习定期预测那些接触新冠肺炎阳性患者的风险，而无需与服务器共享任何设备数据。从这些传感器中提取了几个简单但经过良好验证的参数，以开发COVID风险的本地数字标记。这些创新的核心是事实通过知识提炼和模型压缩为标准联邦学习带来的改进。该项目在亚利桑那州立大学技术办公室的支持下，与行业公司合作，将通过移动应用程序部署和评估FACT，并接触到许多用户。FACT可以扩大一般临床试验中采用的声学措施的临床用途，用于新冠肺炎患者。这项研究还提供了大量的机会，让不同的研究生在确保隐私和安全的同时，能够实现有益于社会的技术挑战。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The spread of the Corona Virus Disease 2019 (COVID-19), a highly-infectious disease caused by a newly discovered coronavirus, has reached pandemic levels across the globe. As the numbers in the USA of infections, critical care interventions and deaths continue to rise, mobile applications (apps) that enable contact tracing (CT) are being rapidly deployed to monitor the spread of COVID-19. However, on-going deployments, based on anonymously sharing tokens without exploiting the rich local device data, are insufficient in monitoring disease spread in a timely manner and are also vulnerable to privacy and security attacks. There is an urgent need to develop CT apps that not only monitor but also intervene to limit COVID-19 spread while respecting user security and privacy. This project addresses this challenge via Federated Analytics based Contact Tracing (FACT), a refined federated learning approach to leverage both device-level data and server capabilities in a private and secure manner. FACT enables prevention and intervention by including hotspot identification, user alerts, and continual assessment of user COVID-19 risk. FACT guarantees a private and secure way to (i) evaluate a user's need for testing or their resilience to exposure, and (ii) assess herd immunity across the population. FACT addresses both the vulnerability of current Bluetooth-based systems to a variety of attacks and limited learning at the server by proposing a secure GPS+Bluetooth system which will enable the server to detect geographical infection clusters in a privacy-preserving manner. FACT also harnesses the rich device-level mobility and acoustic sensing data to periodically predict risks of those exposed to COVID-19 positive patients using federated learning and without sharing any device data with the server. Several simple, but well-validated, parameters are extracted from these sensors to develop local digital markers of COVID risk. At the heart of these innovations are the refinements FACT brings to standard federated learning via knowledge distillation and model compression. This project, with the support of ASU University Technology Office and collaboration with industry companies, will deploy and evaluate FACT via a mobile app and reach many users. FACT can extend the clinical utility of acoustic measures, adopted in general clinical trials, for use in COVID-19 patients. This research also provides immense opportunities to train and expose diverse graduate students to the technical challenges of ensuring privacy and security while simultaneously enabling socially beneficial technologies.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.