NETS: Small: Machine Learning Based Algorithms for Quasi-Static Ad Hoc Wireless Networks

准静态自组织无线网络是全城市网状网、机器对机器控制网络和某些传感器网络的模型，在智能电网、交通电网控制和基础设施监测等应用中具有越来越重要的作用。该项目的目标是设计分布式在线网络协议，学习此类网络的特点并进行自我优化，以实现更可预测的性能，如延迟保证。这对智能电网等网络尤其重要，在这些网络中，端到端延迟必须是可预测的。因子图被用来对允许的网络控制动作的概率分布进行建模。正在研究一种新的信息理论公式，该公式测量协调设备之间和每个设备内的网络堆栈之间的网络操作所需的信息共享。这个问题的解决方案将通过使用机器学习中的统计抽样技术来获得，为网络控制提供在线算法。该算法将是分布式的，因为它是通过最小化所需的信息共享来获得的。因此，该项目将为准静态自组织网络的分布式协议的概率设计提供正式的数学基础，利用信息论中的信息、熵和边信息的概念来量化分布式操作的价值。该项目的成果将在主要的专业会议和期刊上发表和介绍，并将向更广泛的公众开放。该项目将支持在无线网络重要领域培养博士生。所获得的理论框架将在统计工程方法和信息论课程中讨论。利用MAC协议的概念，对基于软件无线电的皮？S本科生实验平台进行扩充，使通信理论专业学生获得系统设计的实践经验。

Quasi-static ad hoc wireless networks are models for city-wide mesh networks, machine-to-machine networks deployed for control, and certain sensor networks, which are of growing importance in applications such as smart electricity grids, transportation grid control and infrastructure monitoring. The goal of this project is to design distributed online network protocols that learn the characteristics of such networks and self-optimize to achieve more predictable performance, such as delay guarantees. This is especially important for networks such as smart grid, where end-to-end delay must be predictable. A factor graph is used to model the probability distribution over the allowed network control actions. A novel information theoretic formulation is being investigated, which measures the information sharing required to coordinate network actions, between devices and across the network stack within each device. The solution to this problem will then be obtained by using statistical sampling techniques from machine learning, providing the online algorithm for network control. This algorithm will be distributed, because it is obtained by minimizing the information sharing required. Thus, the project will provide a formal mathematical basis for the probabilistic design of distributed protocols for quasi-static ad hoc networks, utilizing the information theoretic concepts of information, entropy, and side-information to quantify the value of distributed actions. Results from this project will be published and presented in major professional conferences and journals, and will be available to the wider public. The project will support the training of doctoral students in the important field of wireless networks. The theoretical framework obtained will be discussed in courses on statistical engineering methods and information theory. The MAC protocol concepts will be used to augment the PI?s under-graduate experimental set-up based on software radios, so that students of communication theory can obtain hands-on experience with system design.