CCSS: Online Learning for IoT Monitoring and Management

一些新兴技术进步的核心是物联网（IoT）的概念。从概念上讲，物联网被设想为一个智能网络基础设施，其中大量无处不在的智能设备存在于不同的应用领域，如智能建筑，团体和个性化医疗保健，以及自动驾驶联网汽车，仅举几例。如今，许多物联网应用已经改善了日常生活的许多方面。然而，在充分利用物联网的全部潜力之前，需要解决关键挑战。这反过来又需要创新的机器学习方法，以考虑可扩展性、异质性、适应性和对不可预测的不确定性的鲁棒性——这是新兴物联网监控和管理任务面临的核心挑战。新的算法及其性能需要利用数据科学、优化、统计信号处理、通信和网络方面的最新进展。除了显著影响未来的物联网模块外，从该项目的学习和推理中获得的见解还将为包括智能电网、智能城市和自动驾驶汽车在内的一系列其他领域带来交叉效益。在更广泛的范围内，开发的技术将为基础科学和工程研究提供有价值的工具，并倡导社会拥抱新兴的物联网技术。通过将研究与旨在培养下一代物联网专业人才的教育计划相结合，以及促进学术研究与行业需求的交叉授粉，同时促进和拥抱科学和工程的多样性，将进一步产生更广泛的影响。为了解决核心的物联网挑战，该项目提出了使用专家团队进行实时交互功能学习的基础工具。即使在动态不可预测的环境中，学习算法也将具有自适应性和可量化的性能，而且还具有在以下方面进行扩展的能力：1)物联网中的大量“事物”；Ii)涉及复杂学习任务的高维特征向量；iii)在物联网图上收集、处理和交换的大量数据——这是物联网监控所需要的。学习非线性函数的可扩展性、适应性和鲁棒性优势将进一步渗透到交互式黑盒贝叶斯优化中，以及由专家组成的强化学习——这些优点将提高开环和闭环物联网管理的性能。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。

At the core of several emerging technological advances lies the notion of Internet-of-Things (IoT). Conceptually, IoT is envisioned as an intelligent network infrastructure with a huge number of ubiquitous smart devices present in diverse application domains such as smart buildings, group and personalized healthcare, as well as self-driving connected vehicles, to name a few. Today, a number of IoT applications have already improved many aspects of daily life. However, critical challenges need to be addressed before embracing the full potential of IoT. This in turn calls for innovative machine learning approaches that account for scalability, heterogeneity, adaptivity, and robustness to unpredictable uncertainties -- what are the central challenges facing the emerging IoT monitoring and management tasks. Novel algorithms and their performance need to leverage recent advances in data science, optimization, statistical signal processing, communications, and networking. In addition to markedly influencing future IoT modules, insights gained from this project's learning and inference will also cross-fertilize benefits to a gamut of additional domains, including smart grids, smart cities, and self-driving vehicles. At a broader scale, the developed technologies will provide valuable tools for foundational science and engineering research, and advocate societal embracing of the emergent IoT technologies. Broader impact will be further effected by the integration of research with an educational plan designed to train the new cadre of next-generation of IoT professionals, as well as foster cross-pollination of academic research to industry needs, while promoting and embracing diversity in Science and Engineering.To address the core IoT challenges, this project puts forth foundational tools for real-time interactive function learning using an ensemble of experts. Learning algorithms will be developed with adaptivity and quantifiable performance even in environments with unpredictable dynamics, but also with ability to scale in terms of i) the huge number of ‘Things’ in IoT; ii) the high-dimensional feature vectors involved in sophisticated learning tasks; and iii) the massive data collected, processed, and exchanged over the IoT graph – what is desired for IoT monitoring. Scalability, adaptivity, and robustness benefits in learning nonlinear functions will be further permeated to interactive black-box Bayesian optimization, and reinforcement learning with an ensemble of experts -- merits that will boost performance in open- and closed-loop IoT management.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.