Seth L. Gilbert, National University of Singapore

Title: Collecting Data in a Changing World: Hybrid Aggregation in Mobile Sensor Networks

Abstract:
Consider the problem of studying a city, exploring its evolution over an average week. One might wonder: How do populations move through the city? How do weather and air quality correlate with these movements? Are there patterns to how people dress? Et cetera. To answer these types of questions, imagine equipping a large number of taxicabs with sensors that record everything in their nearby vicinity. The resulting mobile (vehicular) sensor network provides the perfect platform for monitoring and collecting data. Taxis move throughout the city, providing high levels of coverage at a relatively low cost.

In this talk, I will explore several algorithmic issues that arise in collecting and aggregating data in mobile sensor networks. Specifically, I will focus on "hybrid" aggregation protocols where data is first combined in the network before being reported to a central authority (using, for example, the cellular infrastructure). How efficiently can we aggregate the data? How many participating taxis do we need for efficient in-network communication? How do we cope with the constant changes in the vehicular network? Can we leverage hidden patterns in the underlying vehicular movement to improve the efficiency of aggregation? I will address some of these questions, presenting both theoretical results on hybrid aggregation, as well as simulations based on mobility traces from a large set of Singapore taxicabs.