NeTS: Small: Collaborative Research: Fast Online Machine Learning Algorithms for Wireless Networks

除了用于人与人或人对机器通信的传统用途外，无线网络还被设想成为医疗保健、交通和配电等许多新兴应用的支柱。这些应用程序的顺利运行在很大程度上取决于无线网络的令人满意的运行，无线网络负责在代理之间有效和及时地传输信息，尽管突然变化的网络条件和严格的特定于应用程序的服务要求，现有的最先进的无线网络资源分配不能解释应用程序需求的严格和不断变化的要求。本研究建议利用和扩展机器学习的新兴领域，以开发低延迟无线网络的新方法，这对于支持包括低成本医疗保健、节能和安全在内的不同领域的应用的基本服务是必要的。这项研究取得的进展将使整个社会受益，使人们能够在未来高效、低成本地获得这些服务。该项目还将有助于促进未来工程师的培训和教育，为设计高效无线网络算法提供坚实的理论基础和实践考虑。该项目的总体目标是为未来的无线网络开发统一的机器学习和资源分配框架，以适应物理层快速变化的动态和统计数据以及应用层日益严格的服务需求。机器学习的基本问题是在一个随机系统中，当系统背后的统计模型是未知的先验时，如何做出决策。虽然在这个问题上已经有了很多研究，但许多无线网络特有的特征没有被考虑，包括快速变化的网络动态、多个用户之间的交互和依赖关系，以及瞬态延迟性能。该提案的重点是设计快速的在线学习算法，通过考虑无线网络的独特特性，从而显著提高网络性能。

In addition to their traditional use for human-to-human or human-to-machine communication, wireless networks are also envisioned to form the backbone of many emerging applications in health care, transportation, and power distribution. The smooth operation of these applications critically depends on the satisfactory operation of the wireless networks that are responsible for the efficient and timely transfer of information between agents, despite abruptly changing network conditions and stringent application-specific service requirementsExisting state-of-the-art wireless network resource allocation does not account for the stringent and changing requirements of the application demands. This research proposes to leverage and extend the emerging area of machine learning to develop new approaches to low-delay wireless networks that are necessary for the support of essential services with applications in diverse domains including low-cost healthcare, energy savings, and security. Advances made in this research will benefit the society-at-large by enabling efficient and low-cost access to such services in the future. The project will also help advance the training and education of future engineers with a strong foundation on both the theoretical underpinnings and the practical considerations for the design of efficient wireless network algorithms. The broad objective of this project is to develop a unified machine learning and resource allocation framework for future wireless networks that can adapt to rapidly changing dynamics and statistics at the physical layer and the increasingly stringent service requirements at the application layer. The fundamental problem in machine learning is to make decisions in a stochastic system when the statistical model underlying the system is unknown a priori. While there has been much activity on this problem, many features unique to wireless networks are not considered, including rapidly changing network dynamics, interactions and dependencies among multiple users, and transient delay performance. The focus of the proposal is to design fast, online learning algorithms which lead to dramatic improvements in network performance, by taking into account the unique characteristics of wireless networks.