|Stochastic Systems Group|
Professor Samuel Madden
In this talk, I will discuss several problems related to querying sensor networks using probabilistic models. Query processing in wireless sensor networks (WSNs) is different than query processing in traditional databases because data in WSNs does not exist before queries arrive, and must be acquired, usually by the database system. Acquisition costs can consume significant amounts of energy, because of both networking and sensing overheads. Interestingly, there is significant asymmetry in acquisition costs, because some nodes are more network hops away than others, and some sensors are cheaper to sample than others. We can exploit this asymmetry to significantly reduce data acquisition costs by observing low-cost sensors and using these observations, in conjunction with historical observations about correlations between sensor readings, to predict values at of more expensive sensors. I will show how this idea can be used to provide both approximate and exact answers. I will summarize some results in this area and discuss ways in which this work generalizes beyond sensor networks.
Problems with this site should be emailed to firstname.lastname@example.org