Flight over Bristol

Bristol Algorithms Days 2010
Feasibility Workshop

15 - 16 February 2010

  •    •    •  

Adaptive Learning Algorithms for Data Streams

Patrick McSharry

Our ability to model real-world complex dynamical systems has been enhanced by advances in science, wider availability of high quality spatiotemporal observations and increased computational resources. Much of the recent innovation in constructing quantitative models has arisen at the interface of mathematics and computer science. The increasing availability of data streams from a multitude of sources is driving the development of algorithms for supporting decision-makers in many sectors including finance, insurance, and energy. Data streaming offers many challenges and opportunities to those that are able to harness and act upon the additional information content. The development of algorithms for prediction analytics can provide the forecasts, classifications and risk analysis required for improving operational efficiency. These algorithms are constructed to account for uncertainty at all levels of the modelling process and to quantify the risk of extreme events. Ensemble prediction techniques offer a means of quantifying uncertainty and obtaining a probabilistic forecast. Methods for evaluating probabilistic forecasts are discussed. Finally, approaches for communicating information about risk and uncertainty to policy-makers and decision-makers are proposed.


  The University of Bristol   EPSRC