Applications of Probabilistic Constraints

Relational database systems are a successful platform to manage large amounts of data, but do not cope well with uncertainty. However, the amount of uncertain data is growing at an unprecedented rate from both traditional sources (e.g. integrating enterprise data) and from next generation sources (e.g. information extraction). This trend has prompted the database community to investigate a promising new technique, probabilistic databases, that natively handle uncertainty. In this nascent area, it is an open question which techniques from traditional database management apply. A remarkably useful technique in standard relational databases is to allow users to enrich the semantics of their data by declaring constraints. Two traditional uses of constraints are to prevent errors while updating the data and to optimize queries. More recently, constraints provided an elegant solution to the problem of data exchange. These successes give us reason to believe that constraints will play a large role in the theory and implementation of probabilistic databases. This report proposes to generalize constraints to handle uncertainty in the data and the constraints themselves. We identify several traditional and emerging applications that are naturally modeled with probabilistic constraints.