Research project Spectral Subsampling for Large-Scale Bayesian Inference - Theory and Applications
Technological developments have made it possible to collect large datasets over time and space at low cost. The aim of this research project is to develop statistical models and fast estimation methods for such large-scale spatiotemporal data.
Project description
Spatiotemporal data are increasingly common in most scientific fields and industrial sectors; some examples are house prices and consumer purchase data, social networks and mobility flows, travel times on road networks, images of brain activity, sensor data from self-driving cars, meteorological data, and reading habits data from e-books.
The lack of fast statistical algorithms means foregone opportunities to infer complex relationships from large informative datasets that are often not used to anything near their full potential, and model development is lagging behind.
This project will develop new probabilistic models and fast and robust inference methods for statistical models for spatiotemporal problems, with a view toward applications in finance, transportation and neuroscience. Methodologically, this involves a combination of efficient data subsampling and spectral methods for time series and spatial data. Multivariate spatiotemporal data are also a particular focus in the project.
Project members
Project managers
Mattias Villani
Professor
Members
Oskar Gustafsson
Universitetslektor