Our aim is to automate the extraction of knowledge and understanding from data. Allowing machines (and humans) to understand what is happening and to acquire new skills and learn new things. We achieve this by developing new probabilistic models and deriving algorithms capable of learnings these models from data. The systematic use of probability in representing and manipulating these models is key. It allows us to represent not only what we know, but to some extent also what we do not know. We take a particular interest in dynamical phenomena evolving over time.
Our research is multi-disciplinary and it sits somewhere on the intersection of the areas of Machine learning and statistics, signal processing, automatic control and computer vision. We pursue both basic and applied research, which explains our tight collaboration with various companies. A slightly more detailed overview of our research is available here.
Recent research results/news
February 24, 2017 [New research environment will be created] Our research environment “NewLEADS - New Directions in Learning Dynamical Systems" together with researchers at KTH has been granted funding from the Swedish research council. More information is available here.
March 16, 2017 [New online learning method for large/streaming data] We have developed a new online learning method for prediction especially well suited in settings involving large and/or streaming data sets. The predictor is implemented online with a runtime that scales linearly in the number of samples; has a constant memory requirement; avoids local minima problems; and prunes away redundant feature dimensions without relying on restrictive assumptions on the data distribution.
Dave Zachariah, Petre Stoica and Thomas B. Schön. Online learning for distribution-free prediction. Pre-print arXiv:1703.05060. [arXiv]
March 1, 2017 [New results accepted for IFAC World congress 2017] We have some new insights to share on the classic problem of smoothing in linear state space models. More specifically we consider the problem of fixed interval smoothing in linear time-varying Gaussian state space models. Interestingly all existing solutions to this problem impose restrictions on the system matrices in order for them to be applicable. We develop new forward-backward type recursions that are applicable under the mildest assumptions possible.
Li-Hui Geng, Brett Ninness, Adrian G. Wills and Thomas B. Schön. Smoothed state estimation via efficient solution of linear equations. In Proceedings of the 20th World Congress of the International Federation of Automatic Control (IFAC), Toulouse, France, July, 2017.
Click here for older news.