Scalable, interpretable and versatile models of relational data: Design, induction and inference

Collecting, storing, retrieving and analyzing big data poses many scientific and technical challenges. This project focus on analysis, which arguably poses the biggest challenge because even analyzing small amounts of data can be computationally very expensive. From a machine learning perspective, a learned model should satisfy four desiderata. First, scalability, both in terms of learning and use, is paramount. Second, it should be able to explain why a prediction was made. Third, the model should account for the rich, relational structure inherent in many real-world domains. Fourth, it should be versatile: after it is constructed, it can be used for many different purposes. Currently no analysis approach satisfies all these criteria. Therefore, an approach that meets these needs would represent a radically new and very flexible method for performing data analysis. The goal of this project is to develop the foundations of such an approach.

Keywords:relational data

Disciplines:Applied mathematics in specific fields, Computer architecture and networks, Distributed computing, Information sciences, Information systems, Programming languages, Scientific computing, Theoretical computer science, Visual computing, Other information and computing sciences