Fundamental Bounds on Decentralized Adaptive Detection in Hidden Markov Models

摘要为？TF08：隐马尔可夫模型中分散自适应检测的基本边界？（PI: Yajun Mei, NSF proposal #0830472）在现代信息时代，传感器、计算和通信技术的进步为决策者和组织在传感器网络系统的许多实际应用领域快速做出有效决策提供了有希望的机会。然而，如果不及时更新或适应不断变化的环境，即使是最好的决策方法在生物恐怖主义和黑客攻击等威胁检测应用中也不可避免地脆弱。由于传感器之间复杂的时空相关性以及通信、能源和计算方面的限制，挑战变得更加困难。本研究关注的是当传感器观测来自隐马尔可夫模型时，分散自适应检测的一般和系统基础和方法的发展，重点是推导可靠检测变化能力的基本信息限制。针对传感器网络系统的两种情况，研究了隐马尔可夫模型中的分散自适应检测。第一种是传感器无法访问其过去观测的系统，其中研究主题包括(i)最优平稳量化器；（ii）自适应检测下限；（iii）通过串联量化器进行鲁棒检测；（iv）剔除传感器观测值的自适应检测。第二种是传感器可以访问其过去观察的系统，其中研究调查：(i)通用信息界限；（ii）方便计算的有效方案；（iii）检测延迟可控的方案；（四）分组传输。

Abstract for ?TF08: Fundamental Bounds on Decentralized Adaptive Detection in Hidden Markov Models? (PI: Yajun Mei, NSF proposal #0830472)In modern information era, the advance of sensor, computing and communication technologies offers promising opportunities for the decision makers and organizations to make effective decisions quickly in many areas of real-world applications of sensor network systems. However, without timely updating or adaptation to reflect the changing environments, even the best decision-making methods are irrevocably vulnerable in applications such as threats detection in bioterrorism and hacking. The challenges become more difficult due to the complex spatio-temporal correlations among sensors and the constraints on communications, energy and computing. This research is concerned with the development of a general and systematic foundation and methodologies for decentralized adaptive detection when sensor observations are from hidden Markov models, with the focus on deriving the fundamental information limitation on the ability to reliably detect the changes. The investigator studies decentralized adaptive detection in hidden Markov models for two scenarios of sensor network systems. The first is the system where sensors do not have access to their past observations, in which the research topics include (i) optimal stationary quantizers; (ii) lower bounds on adaptive detection; (iii) robust detection via tandem quantizers; and (iv)adaptive detection with censored sensor observations. The second is the system where sensors have access their past observations, in which the research investigates: (i) universal information bounds; (ii) computing-friendly, effective schemes; (iii) schemes with controlled detection delay; and (iv) blockwise transmission.