Postprint version. Published in International Conference on Database and Expert Systems Applications (DEXA 2003) Proceedings: Prague, Czech Republic, August 1, 2003, pages 527-536.
NOTE: At the time of publication, the author Alex Dekhtyar was not yet affiliated with Cal Poly.
The groundswell for the `00s is imprecise probabilities. Whether the numbers represent the probable location of a GPS device at its next sounding, the inherent uncertainty of an individual expert's probability prediction, or the range of values derived from the fusion of sensor data, probability intervals became an important way of representing uncertainty. However, until recently, there has been no robust support for storage and management of imprecise probabilities. In this paper, we define the semantics of traditional query algebra operations of selection, projection, Cartesian product and join, as well as an operation of conditionalization, specific to probabilistic databases. We provide efficient methods for computing the results of these operations and show how they conform to probability theory.