EAGER: Collaborative Research: Understanding Human Behaviors and Mental Health using Federated Machine Learning on Smart Phones

了解人类行为和心理健康对现代社会变得越来越重要。正在爆发的冠状病毒（COVID-19）不仅进一步凸显了其重要性，而且要求立即采取行动。该项目将在移动设备上开发一个联邦机器学习（FL）框架和应用程序，以了解人类行为和心理健康。计划中的研究将协同跨学科研究，特别是推动联邦学习和公共卫生的信封。该项目不仅将为联邦学习社区提供重要而及时的现实应用、健康行为监测和预测，还将以独特而详细的角度促进我们对移动设备的身心健康以及COVID-19对人类社会的影响的理解。本项目将跨学科的研究成果整合到课程中，培养来自弱势群体的学生。从技术上讲，该项目有两个主要组成部分：1)数据收集和统计分析；2)构建联邦学习框架和应用程序。在第一部分中，该项目将从城市和郊区的学生亚群中收集基于智能手机的传感器数据，以及其他与健康相关的调查和数据。该项目将具体分析和确定从手机收集的哪些数据可以作为行为和心理健康的指标和因果因素。在第二部分，该项目将开发深度学习模型，在严格的隐私保护下，预测人类行为、身心健康状况/趋势。具体来说，预测模型将在联邦学习设置中开发，以在具有差分隐私保证的设备上本地训练模型，而无需将传感器数据收集到云端。最后，该项目将在移动电话上开发一个基于联邦学习的行为监测和预测应用程序，并将根据第一个组件的队列研究评估原型系统。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。

Understanding human behaviors and mental health are becoming increasingly important for modern society. The ongoing outbreak of coronavirus (COVID-19) not only further highlights its importance but also calls for immediate action. This project will develop a federated machine-learning (FL) framework and application on mobile device for understanding human behaviors and mental health. The planned research will synergize interdisciplinary research and particularly push the envelopes of federated learning and public health. This project will not only provide an important and timely real-world application, health-behavior monitoring and prediction, for the federated learning community, but also will advance our understanding of physical and mental health through mobile devices, and the impacts of COVID-19 to human society in a unique and detailed angle. This project will integrate the interdisciplinary research results into courses, and train students from underrepresented groups. Technically, the project has two main components: 1) Data collection and statistical analysis, and 2) Building federated learning framework and application. In the first component, the project will collect smartphone-based sensor data from student sub-population in both urban and suburban areas along with other health related surveys and data. The project will specifically analyze and determine what data collected from the mobile phone can be the indictors and causal factors of behavior and mental health. In the second component, the project will develop deep learning models to predict human behaviors, physical and mental health conditions/trends over time, under rigorous privacy protection. Specifically, the prediction models will be developed in federated learning settings to train the model locally on the device with differential privacy guarantees, without collecting sensor data to the cloud. Finally, the project will develop a federated learning based behavior monitoring and prediction application on mobile phones and will evaluate the prototype system on the cohort of studies from first component.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.