Title: Optimal Monitoring of Sensor Networks
Funding source: AFOSR/DOD (Air Force Office of Scientific Research / Department of Defense)
Sensor networks form a key part of the growing infrastructure for making informed decisions concerning available and potential data.
Data can take many forms, and in all cases, a potentially massive amount of data can be collected and then used to make all types of decisions. For example, telemetry data is used in guidance systems, environmental data in policy decisions, geo-statistical in early-warning systems, and financial data in investment decisions.
Networks have limited bandwidth, and there are enormous computational challenges for rapidly processing data and even storing it. Because of these challenges, it is very important to identify the most important sources/locations for monitoring, so that we can make optimal decisions, often in nearly real time. The real-time interest is because as conditions change, we have to make decisions that are relevant to the current state of sometimes rapidly changing systems.
The problem of determining the optimal set of locations for sensors is a very challenging problem for modeling and computation. We have to account for the complex interactions of multivariate data per sensor, and the complicated structure of relations between nearby sensors. Perhaps the greatest related challenge is that we want to design sensor networks so that the data collected is adequate across a wide variety of uses. Because of this last point, we consider an information-theoretic measure of information, rather than a limited-use parametric statistical model.
Our technical approach is based on advanced techniques of mathematical optimization and data analysis. In that context, there is no standard algorithm for handling our problem. So we are developing new techniques to account for the complex nature of our formulation.
We anticipate being able to solve larger and more realistic instances of sensor-network optimization problems than the current state-of-the-art, with the potential for impact across many domains.