ProDIS · Provenance for Data-Intensive Systems

In the context of data-intensive systems, data provenance captures the way in which data is used, combinedand manipulated by the system. Provenance information can for instance be used to reveal whetherdata was illegitimately used, to reason about hypothetical data modifications, to assess the trustworthinessof a computation result, or to explain the rationale underlying the computation.As data-intensive systems constantly grow in use, in complexity and in the size of data they manipulate,provenance tracking becomes of paramount importance. In its absence, it is next to impossible to follow theflow of data through the system. This in turn is extremely harmful for the quality of results, for enforcingpolicies, and for the public trust in the systems.Despite important advancements in research on data provenance, and its possible revolutionary impact,it is unfortunately uncommon for practical data-intensive systems to support provenance tracking. Thegoal of the proposed research is to develop models, algorithms and tools that facilitate provenancetracking for a wide range of data-intensive systems, that can be applied to large-scale data analytics,allowing to explain and reason about the computation that took place.Towards this goal, we will address the following main objectives: (1) supporting provenance for moderndata analytics frameworks such as data exploration and data science, (2) overcoming the computationaloverhead incurred by provenance tracking, (3) the development of user-friendly, provenance-based analysistools and (4) experimental validation based on the development of prototype tools and benchmarks.