Distributed Methods for Statistical Decision Making in Networked Environments

摘要：传感器和计算技术的最新进展使得大规模分布式传感器网络能够作为候选技术在诸如建筑安全、环境感测、栖息地监测、电力系统和制造等各种应用中提供实时信息。 这种设备的网络传感的潜力是巨大的，但有显着的系统级的挑战，实现这一潜力的设计是可靠的，高效的和可扩展的。本研究涉及的发展的基本系统级框架的不确定性决策在ad-hoc网络和功率有限的环境。虽然这些问题中的一些已经beenpreviously在隔离处理，我们将开发一个统一的方法，解决多个相互作用的依赖关系，以获得系统级的工具，合理的设计和实施。因此，本研究有三个主要的推力：1）自适应查询策略（AQS）-一个自上而下的观点，基于层次结构，涉及指定的决策代理。 在这方面，我们提出了新的反馈策略，传感器查询在低数据速率（低粒度）。 2)分布式检测--一种自下而上的方法，通过在子网络上传递本地消息来达成共识。 3)分布式估计--提出了在操作约束下从一组已知传感器到已知目的地的信息粒度缩放策略。这种方法的新奇在于开发用于传感器融合的工具，用于传输在随机噪声中观察到的非遍历、非随机参数的信息。

Abstract: Recent advances in sensor and computing technologies enable massively distributed networks of sensors as candidate technologies to provide real-time information in diverse applications such as building safety, environmental sensing, habitat monitoring, power systems and manufacturing. The potential of networked sensing by such devices is enormous, yet there are significant system-level challenges in realizing this potential by designs that are reliable, efficient and scalable.This research involves the development of fundamental systems-level framework for uncertain decision making in ad-hoc networked and power limited environments. Although some of these issues have beenpreviously tackled in isolation, we will develop a unified approach, addressing multiple interacting dependencies to obtain systems level tools for rational design and implementation. Accordingly this research has three principle thrusts: 1) Adaptive Querying Strategies (AQS) -- a top-down viewpoint based on hierarchical architectures that involve designated decision agents(s). In this thrust we propose novel feedback strategies for sensor querying at low data rates (low granularity). 2) Distributed Detection -- a bottom-up approach which amounts to consensus building via local message passing on sub-networks. 3) Distributed Estimation -- Strategies for scaling information granularity from a set of known sensors to known destinations under operational constraints is proposed. The novelty of this approach is to develop tools for sensor fusion for transmitting information about non-ergodic, non-random parameters observed in stochastic noise.