CAREER: Ubiquitous and Time-Critical Federated Learning with Cooperative Mobile Edge Networking

联邦学习（FL）使网络边缘的物联网（IoT）设备能够协作学习共享预测模型，同时将所有个人数据保存在设备上。但目前基于云的FL由于物联网设备与云之间传输距离较长，无法满足延迟敏感型物联网应用的时延需求。该项目旨在在无线边缘实现无处不在和时间关键的FL，以支持延迟敏感和数据驱动的物联网应用。该项目将满足许多具有重大经济和社会影响的引人注目的应用的需求，如增强现实、自动驾驶、移动医疗保健和智能制造。该项目的教育议程包括向K-12高中教师提供教育夏令营，指导本科生和研究生，特别是少数民族和代表性不足的群体进行研究，并通过新课程开发和研讨会向学生和行业合作伙伴传播研究成果。本项目开发了一种基于协同移动边缘网络的新型联邦学习（FL）框架，该框架可以高效地支持分布式物联网（IoT）数据的学习和决策，具有高精度、低延迟和保证隐私的特点。本项目研究了三个相互关联的研究重点：1)在两级网络结构下设计新颖的网络感知学习算法，以确保无线边缘网络上物联网设备上分散数据的高效和有效的模型训练；2)基于深度强化学习联合优化资源配置和学习，在系统异构和资源约束下快速学习出准确的模型；3)开发新颖的差分隐私技术，严格保护物联网设备上个人数据的隐私，同时保持较高的模型准确性和降低通信成本。拟议的研究将使下一代无线边缘网络能够支持大量延迟敏感和数据驱动的物联网应用。通过弥合不断发展的移动计算和网络技术与快速发展的机器学习技术之间的差距，拟议的研究将不仅有利于无线网络，而且有利于机器学习研究社区。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。

Federated learning (FL) enables Internet-of-Things (IoT) devices at the network edge to collaboratively learn a shared prediction model while keeping all personal data on the device. However, the current cloud-based FL fails to meet the latency requirements of delay-sensitive IoT applications due to the long-distance transmission between IoT devices and the cloud. This project aims to enable ubiquitous and time-critical FL at the wireless edge to support delay-sensitive and data-driven IoT applications. The project will fulfill the needs of many compelling applications with significant economic and societal impacts such as augmented reality, autonomous driving, mobile healthcare, and smart manufacturing. The project’s educational agenda includes outreach to K-12 with educational summer camps for high-school teachers, mentoring undergraduate and graduate students, especially from minority and underrepresented groups, in the research, and disseminating research outcomes to students and industry partners through new course development and seminars.This project develops a novel Federated learning (FL) framework based on cooperative mobile edge networking that can efficiently support learning and decision making on distributed Internet-of-Things (IoT) data with high accuracy, low latency, and guaranteed privacy. Three interconnected research thrusts are investigated in this project: 1) design of novel network-aware learning algorithms under a two-level network structure to ensure efficient and effective model training from decentralized data on IoT devices over wireless edge networks; 2) jointly optimize resource allocation and learning based on deep reinforcement learning to learn an accurate model rapidly under system heterogeneity and resource constraints; 3) develop novel differential privacy techniques to rigorously protect the privacy of personal data on IoT devices while maintaining high model accuracy and reducing communication cost. The proposed research will enable next-generation wireless edge networks that support a plethora of delay-sensitive and data-driven IoT applications. The proposed research will benefit not only the wireless networking but also machine learning research communities by bridging the gap between the evolving mobile computing and networking technologies and rapidly advancing machine learning techniques.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.