CRII: CNS: IoT-aware Federated On-Device Intelligence

该奖项全部或部分由《2021年美国救援计划法案》（公法117-2）资助。设备上机器学习的最新突破使人工智能（AI）更接近物联网（IoT）设备，将物联网范式从“连接的事物”转变为“连接的智能”。考虑到机密物联网数据的存在和数据泄露的广泛影响，联邦学习提供了一种保护隐私的解决方案，通过交换设备上的模型更新而不是私有物联网数据来实现知识共享。然而，经典的联邦学习假设具有丰富标记数据的同质参与设备，这与大多数物联网设备的属性不一致，例如资源约束、异构性和缺乏注释。为了释放物联网系统中联合设备智能的潜力，该项目侧重于两个关键但尚未解决的问题：(i)以无数据的方式跨资源受限和异构物联网设备进行联合知识共享；（ii）在不断变化的物联网环境下，使用有限的地面真理标记知识进行联合领域适应。该项目开发了新颖实用的方法，以解决知识生成和转移阶段在准确性、效率和数据依赖性方面的冲突目标。所提出的研究将在模拟器驱动和现实世界的测试台上进行彻底和严格的评估。该项目将推进目前对实际但具有挑战性的物联网环境中联邦学习的理解，解决联邦设备上智能广泛参与的挑战，并通过联邦智能实现引人注目的新物联网应用。该项目的研究成果将与现有课程和K-12课程相结合，并通过会议、研讨会和出版物传播，以加速人工智能和物联网研究的进展。此外，该项目将积极参与本科生和弱势群体学生，并改善弱势群体在计算机科学和工程研究中的存在。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。

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2).Recent breakthroughs in on-device machine learning bring artificial intelligence (AI) closer to Internet-of-Things (IoT) devices, shifting the IoT paradigm from “connected things” to “connected intelligence.” Given the presence of confidential IoT data and the widespread impact of data breaches, federated learning provides a privacy-preserving solution that enables knowledge sharing by exchanging on-device model updates rather than private IoT data. However, classical federated learning assumes homogeneous participating devices with a wealth of labeled data, which stands at odds with the properties of most IoT devices, such as resource constraints, heterogeneity, and lack of annotation. To unleash the potential of federated on-device intelligence in IoT systems, this project focuses on two critical yet open problems: (i) federated knowledge sharing across resource-constrained and heterogeneous IoT devices in a data-free manner; and (ii) federated domain adaptation with limited ground truth labeled knowledge under ever-changing IoT environments. The project develops novel and practical approaches to address conflicting goals on accuracy, efficiency, and data dependence at both the knowledge generation and transfer stages. The proposed research will be thoroughly and rigorously evaluated in simulator-driven and real-world testbeds.This project will advance the current understanding of federated learning in practical yet challenging IoT environments, address challenges to broad participation of federated on-device intelligence, and enable compelling new IoT applications with federated intelligence. The research outcomes of this project will be integrated with existing curriculums and K-12 programs and disseminated through conferences, seminars, and publications to accelerate progress in AI and IoT research. Furthermore, this project will actively involve undergraduate and underrepresented students and improve the presence of underrepresented minorities in computer science and engineering research.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.