Abstract: Wireless sensor networks stand to enable data collection from a multitude of environments on unprecedented scales. Network nodes are energy-constrained, with radio communication dominating energy consumption. The naive method of periodically reporting all data to a base station for centralized processing is unacceptable due to its high message traffic.

I will present model-encoding suppression schemes for producing continuous query results without continuous reporting. The most effective models incorporate both temporal and spatial correlations. While building schemes to leverage either of these types is straightforward, combining them is not. I will discuss a scheme, Conch, which effectively couples the correlations by temporally monitoring spatial constraints, to minimize transmission cost.

Sensor networks are prone to transmission failures, which are especially detrimental to suppression schemes; a suppressed report cannot be distinguished from a failed one. The cost of implementing reliable transmission in the communication layer is prohibitively high. Instead, I show how to make schemes robust with application-level redundancy in monitoring and reporting. I discuss BaySail, a framework that uses Bayesian analysis to recover missing data. In inference, it augments data received from the network with knowledge of the suppression scheme and the spatio-temporal correlations to highly constrain the uncertainty caused by missing data.

Short Bio: Adam Silberstein is graduating with a Ph.D. from the Computer Science Department at Duke University. His research is in data management, with specific interests in sensor network query processing and XML. He also generally interested in sensor network protocols. Prior to entering graduate school, he graduated from Duke University with a B.A. in Computer Science and English.