The HeisenData Project (HeisenData)
Eirinaios Chrysovalantis Michelakis, Zhe Daisy Wang, Joseph M. Hellerstein, Michael Franklin and Minos Garofalakis
The convergence of embedded sensing and stream query possessing suggests an important role for database techniques in managing data that only partially--and often inaccurately--capture the state of the world. Reasoning about uncertainty inside a database management system as a first class citizen becomes an increasingly important operation that is needed to provide support for a variety of emerging applications in the context of sensor monitoring and pervasive computing. Existing approaches of "injecting" probabilistic semantics into the relational world explored rather simple uncertainty models, in which data were treated independently and often at the wrong semantic granularity.
The HeisenData Project attempts to integrate the well-understood deterministic data management framework with sound probabilistic reasoning by providing the ability to learn, inference, and query a variety of realistic probabilistic graphical models that capture dependencies among the data stored in the database. This functionality can be incorporated into the existing data processing framework by adding new probabilistic operators to the query processing and optimization engine. Furthermore, the automatic construction and maintenance of the probabilistic models arises as an important problem that gets exaggerated as the scale of the data and query workload served increases.