Feasible Algorithms for Big Inference

Abstract:: Many tried-and-true tools of statistical inference are infeasible in the Big Data context due to computational bottlenecks. This project will develop new algorithms for computationally-intensive statistical tools, which will allow classically-trained scientists to analyse Big Data. Such tools are false discovery rate control, heteroscedastic and robust regression, and mixture models. This will be achieved via new developments in Big Data-appropriate optimisation and composite-likelihood estimation. Open, well-documented, and accessible software will be made available for the scalable and distributable analysis of Big Data. The outcome will be a suite of new scalable algorithms for scientist conducting inference within the Big Data context.
Grant type:: ARC Discovery Early Career Researcher Award
Funded by:: Australian Research Council

The University of Queensland