| Abstract |
15 years ago, Tim-Berners Lee, the inventor of the Web, envisioned an Oh, Yeah? button that when clicked would provide meta-information (i.e. the provenance) about why a user should trust a Web page. This project’s ambitious aim is to create an Oh, Yeah? button for all of our data.
The need is clear. Decision makers are faced with overwhelming amounts of data (9.5 zettabytes a year) that is analyzed using complex and opaque processes. They do not know how, by whom, or why data was produced and have to simply rely on its quality and trustworthiness. Currently, capturing provenance relies on modifying software, which must be installed across the data-production pipeline before provenance is needed. Thus, I pose the research question: how can we automatically reconstruct the provenance of data from the computational environment in which it resides? The two keywords in this question are “automatically” (i.e. without expensive human intervention), and “re-construct” (i.e. not relying on up-front modification of software.) I attack the problem with advances in two areas: provenance representation and abductive reasoning.
Note: this is a practice talk for an ERC starting grant interview
|