CAREER: Distributed Intelligence in Future Wireless Networks: From System-Aware Learning to System-Learning Co-designs

鉴于无线设备的传感和计算能力不断增强，再加上对数据隐私的担忧日益加剧，大量人工智能和机器学习任务正在无线网络中的设备上运行。不幸的是，大多数学习算法的发展很少关注底层系统约束；而且，大多数现有的通信和网络设计很少考虑到运行学习任务的独特特征。为了促进快速机器学习发展和无线网络设计之间的协同作用，该项目旨在实现分布式学习算法和无线网络的变革性协同设计。这个CAREER项目将进一步整合教育计划和研究目标，i)用分布式学习组件改造现有的传感器网络课程；Ii)直接让本科生和研究生参与研究，特别是来自代表性不足群体的学生；iii)向公众，特别是K-12学生和教师拓展。为了实现这个目标，这个CAREER项目将追求i)系统感知学习，以及ii)系统学习协同设计。对于系统感知学习，该项目将首先开发资源高效的分布式学习算法，从而学习更新将被简约地执行。将研究这些资源高效算法的鲁棒实现，以考虑用户移动性和不可靠无线信道中的对抗性攻击。在系统学习协同设计方面，该项目将开发新的边学习边管理算法，通过联合学习、功率控制、排队和工作量管理方案，最大限度地提高学习精度。该项目提出了一个雄心勃勃的计划，以实现未来无线网络的系统学习协同设计。它的基本进步也将影响社会技术系统，如电网、城市交通系统和水/气分配系统。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。

Given the growing sensing and computing power of wireless devices, amplified by increasing concerns on data privacy, a sizeable number of artificial intelligence and machine learning tasks are running at the devices distributed in wireless networks. Unfortunately, most of learning algorithm developments pay little attention to the underlying system constraints; and, most of existing communication and network designs rarely account for the unique characteristics of running learning tasks. To catalyze the synergies between rapid machine learning developments and the wireless network designs, this project aims at a transformative co-design of distributed learning algorithms and wireless networks. This CAREER project will further integrate an educational plan with the research goals by i) revamping the existing sensor network course with distributed learning components; ii) directly involving undergraduate and graduate students in research, especially from under-represented groups; and, iii) outreaching to the general public, in particular K-12 students and teachers.Towards this goal, this CAREER project will pursue i) system-aware learning, and ii) system-learning co-designs. For system-aware learning, the project will first develop resource-efficient distributed learning algorithms, whereby learning updates will be executed parsimoniously. Robust implementation of these resource-efficient algorithms will be studied to account for user mobility and adversarial attacks in unreliable wireless channels. Regarding system-learning co-designs, the project will develop new learning-while-managing algorithms that maximize the learning accuracy through joint learning, power control, queueing, and workload management schemes. This project presents an ambitious plan to enable system-learning co-designs of future wireless networks. Its fundamental advances will also impact sociotechnical systems, such as power grids, urban transportation systems and water/gas distribution systems.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.