A Systems Approach to Networked Decision Making in Uncertain Environments This proposal concerns the development of tools for a fundamental system-level framework for networked decision making in uncertain environments. While significant effort over the last decade in sensor development, physical layer transmission and networking has laid the initial groundwork for practical deployment, the full potential for networked sensing systems can only be realized through a fundamental understanding of decision-making in networked and uncertain environments. In keeping with this view, the proposed research will represent communications and networking aspects as mathematical constraints and develop methods for a distributed group of decision agents to reliably detect, localize, and track relevant dynamic and uncertain events under these constraints. Intellectual Merit: We organize our proposed research into four thrusts based on two funda- mentally different types of problems that arise in many (environmental) applications: (a) where each networked sensor/decision agent observes part of a global phenomena; (b)a local phenomena observable only by a small number of sensors. While the former case requires sensor collabora- tion across space, the latter requires rapidly searching for sensors with desired information in a decentralized manner. The four major thrusts are: Consensus Based Approach for Networked Decision Making: This thrust is concerned with distributed inference when observations of global phenomena are involved. The approach amounts to local information refinement followed by local message passing. This is both efficient and practically appealing. The solution techniques rely on properties of dynamical systems whose dynamics are characterized by network connectivity. Adaptive Decentralized Sampling: Here we address the localized information case. We pro- pose a feedback perspective that is inspired by recent developments in testing large number of hypothesis in the statistics literature. Estimation under communication constraints: Here motivated by applications we develop mixed stochastic/deterministic methods for communicating non-ergodic, non-random param- eters observed in stochastic noise. Dynamical Scenarios: In this thrust control theoretic methods for networked decision making in highly dynamic scenarios is developed. Broad Impact: The proposed research has the potential to broadly impact existing knowledge in the fields of networking, information theory, estimation, control, and signal processing as well as the broader engineering community. In addition to disseminating our results through publications in scholarly journals, conference presentations and on-line we envision the following outcomes: Societal: Successful completion of the program will shed light on the way sensor networks are configured and operated for reliable, seamless and prolonged operation. The program is therefore expected to have societal impacts through the diverse applications such as emergency relief services, security/surveillance and other applications that closely relate to public welfare. Education: We plan on introducing a first-year graduate level course that will seek an integrated viewpoint of control, networks and information. 1
