Adrian Onet

Closed world chasing

We give a new closed world semantics for data exchange called constructible solutions semantics, and argue that this semantics is well suited for answering non-monotonic queries in data exchange. We show that the space of constructible solutions can be represented by conditional tables obtainable trough a novel conditional chase procedure which generalizes the classical chase.

Data correspondence, exchange and repair

Checking the correspondence between two or more database instances and enforcing it is a procedure widely used in practice without however having been explored from a theoretical perspective. In this paper we formally introduce the data correspondence setting and its associated computational problems: checking the existence of solutions, and verifying whether a candidate is a solution, for three distinct types of solutions. Data correspondence is a generalization of data exchange and peer data exchange, and a special case of repairing inconsistent databases. We introduce a new class of dependencies, called semi-LAV, that properly includes both LAV and full dependencies, while retaining tractability for peer data exchange, data correspondence, and database repairs. We also introduce the concept of Σ-satisfying homomorphisms. This new type of homomorphism, together with recent advances, is essential in achieving tractability, while at the same time allowing a large class of dependencies, namely the aforementioned semi-LAV ones. We also show the intractability for a series of problems in the case of weakly acyclic tuple generating dependencies. This implies that many tractability results for weakly acyclic dependencies do not carry over to data correspondence; in these new settings we need to restrict the dependencies a bit further, yielding our semi-LAV dependencies.

Higher Order Mining from Sources with Different Competences

The value of data mining on large quantities of data is well known. However there are cases when we can’t access directly the raw data, such as: (i) institutions interested in sharing knowledge may not be allowed to share the raw data; (ii) data is in form of streams and it is only temporarily available for processing; (iii) finally there may also be limits on the computation speed that could be achieved. Therefore the data must be summarized to be processed efficiently. In this paper we consider mining patterns retrieved from sources with different competences where it is needed to select only the “most” supported knowledge trough the sources. The method proposed is based on the knowledge uniformization and apply a common algorithm that will select the most supported knowledge taking into account the competency of the sources.

PossDB: An Uncertainty Database Management System

Management of uncertain and imprecise data has long been recognized as an important direction of research in data bases. With the tremendous growth of information stored and shared over the Internet, and the introduction of new technologies able to capture and transmit information, it has become increasingly important for Data Base Management Systems to be able to handle uncertain and probabilistic data. As a consequence, there has lately been significant efforts by the database research community to develop new systems able to deal with uncertainty, either by annotating values with probabilistic measures or defining new structures capable of capturing missing information (e.g. Trio [3] and MayBMS [2]).

Mining from Incomplete Patterns

Often in higher order data mining (HOM) applications we need to use incomplete patterns received from different sources. In this paper, a solution is proposed for mining new patterns from incomplete or corrupted patterns received from the sources. The solution is based on a relation representation for the patterns, this representation may be used both for pattern discovery from incomplete information, and also for higher order data mining from sources with different competences