Collaborative Research: SHF: Small: Artificial Intelligence of Things (AIoT): Theory, Architecture, and Algorithms

人工智能和物联网的融合创造了物联网智能（AIoT），预计这不仅将提高终端设备的智能，而且还将更好更快地释放物联网数据的力量。鉴于许多领域都存在机密和分布式物联网数据，联邦学习是一种很有前途的方法，可以通过实现协作智能而无需将私有终端设备数据迁移到中央服务器来释放AIoT的潜力。然而，最先进的人工智能对存储和计算资源的沉重负担与大多数资源受限的物联网硬件平台不一致，这在AIoT中部署联邦智能时带来了艰巨的挑战。该研究团队探索硬件高效的人工智能技术，以支持跨不同物联网硬件平台的联合知识转移，从而从理论、架构和算法角度扩展AIoT的范围。拟议的研究为采用人工智能和物联网技术的广泛学科带来了实实在在的好处，促进了人工智能和物联网的融合。该项目为代表性不足群体的本科生和研究生提供培训机会。AIoT主题和研究成果的推广工作针对K-12受众。该项目提供了理论和经验证据，以促进在联合物联网环境中部署硬件高效的人工智能技术，这填补了一个关键的空白-现有方法无法解决AIoT中广泛的资源，效率和隐私挑战。该项目包括四个方面：（1）从显微镜操作，神经量化，实现硬件高效的AI，以理论上指导跨各种物联网硬件平台的联邦智能的专门量化，（2）探索另一种突破性的硬件高效的AI技术，神经架构修剪，以数据不可知的方式寻求最佳的子网络架构，（3）识别新的隐私漏洞，并为AIoT设计开发防御机制，以鼓励广泛参与，（4）建立通用AIoT测试床。通过架构-算法-硬件协同设计，该研究旨在释放各种物联网硬件平台和联邦智能的最大潜力，以扩大AIoT应用范围。该奖项反映了NSF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力评审标准进行评估，被认为值得支持。

The fusion of AI and IoT creates Artificial-Intelligence-of-Things (AIoT), which is expected to not only boost the intelligence on end devices, but also unleash the power of IoT data better and faster. Given the presence of confidential and distributed IoT data in many fields, federated learning has been one promising approach to unlock the potential of AIoT by enabling collaborative intelligence without migrating private end-device data to a central server. However, the heavy burden of state-of-the-art AI on storage and computing resources stands at odds with most IoT hardware platforms that are resource-constrained, which raises daunting challenges when deploying federated intelligence in AIoT. The research team explores hardware-efficient AI techniques to support federated knowledge transfer across diverse IoT hardware platforms to expand the scope of AIoT from theory, architecture, and algorithm perspectives. The proposed research brings tangible benefits to a broad range of disciplines that employ AI and IoT technologies, promoting the fusion of AI and IoT. The project provides training opportunities for undergraduate and graduate students from underrepresented groups. The outreach efforts on AIoT topics and research findings are directed towards K-12 audiences. This project provides the theoretical and empirical evidence to facilitate the deployment of hardware-efficient AI techniques in federated IoT environments, which fills a critical void - the existing approaches fail to address the widespread resource, efficiency, and privacy challenges in AIoT. This project consists of four aspects: (1) enabling hardware-efficient AI from microscope operations, neural quantization, to theoretically guide specialized quantization for federated intelligence across various IoT hardware platforms, (2) exploring another ground-breaking hardware-efficient AI technique, neural architecture pruning, to seek optimal sub-network architectures in a data-agnostic manner, (3) identifying new privacy vulnerabilities and developing defensive mechanisms for the AIoT designs to encourage broad participation, (4) establishing a general-purpose AIoT testbed. Through the architecture-algorithm-hardware co-design, the research intends to unleash the utmost potential of various IoT hardware platforms and federated intelligence to expand the scope of AIoT applications.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.