CICI: Data Provenance: Protecting Provenance Integrity and Privacy

Sri International, Menlo Park CA

Investigators

Abstract

The amount of data associated with scientific and medical analysis has grown, and in parallel, the desire to share resulting data has increased. Similarly, as more science and medical analysis is performed by exploring large volumes of data, with numerous transformations along the way, tracking the history of a result is of increasing utility. Despite these incentives, concerns such as respecting human subject and patient privacy, maintaining intellectual property, and compliance with copyright laws remain as barriers to sharing data with its detailed provenance. By simplifying the process of safely releasing provenance metadata, the project will facilitate the sharing of data with a concomitant benefit to science and medicine. The project creates software to tightly couple data and its accompanying provenance metadata, and sanitize the resulting provenance when it is shared. A storage overlay is developed to perform partial computation over the data as it enters the system. This allows the application of pre-configured functions to incoming data without affecting extant workload performance, by leveraging idle processing resources. The paradigm facilitates efficient content-based provenance that describes the state of the data in the system. This allows references in the provenance metadata to be strongly linked to integrity claims about the specific data inputs and outputs, even when these correspond to large data sets. When such scientific and medical data sets are combined, new privacy-violating inferences may be drawn. The project investigates techniques that operate on sets of provenance graphs, descriptions of unsafe inferences, and ontological or probabilistic characterizations of the relationships of provenance elements. A tool is developed to transform the provenance into graphs with reduced leakage of sensitive information that are still usable for pre-defined provenance queries, and to identify which elements need auxiliary protection.