Management of data with uncertainties

Probabilistic databases, Query processing Since their invention in the early 70s, relational databases have been deterministic. They were designed to support applications s.a. accounting, inventory, customer care, and manufacturing, and these applications require a precise semantics. Thus, database systems are deterministic. A row is either in the database or is not; a tuple is either in the query answer or is not. The foundations of query processing and the tools that exists today for managing data rely fundamentally on the assumption that the data is deterministic. Increasingly, today we need to manage data that is uncertain. The uncertainty can be in the data itself, in the schema, in the mapping between different data instances, or in the user query. We find increasingly large amounts of uncertain data in a variety of domains: in data integration, in scientific data, in information extracted automatically from text, in data from the physical world. Large enterprises today can sometimes afford to cope with the uncertainty in their data by completely removing it, by using some expensive data cleaning or ETL tools. But increasingly today organizations or users need to cope directly with uncertain data, either because cleaning it is prohibitively expensive (e.g. in scientific data integration or in integration of Web data), or because it is even impossible to clean (e.g. sensor data or RFID data). It becomes clear that we need