SC²S Colloquium - March 5, 2014
|Date:||March 5, 2014|
|Time:||3 pm, s.t.|
Michael Lettrich: Dimension- and Local-Adaptive Refinement Strategies for Data Mining Using Sparse Grids
Regression on high dimensional data sets is a complicated problem. Most simple algorithms can not cope with the huge amounts of data, spread over various dimension. This talk discusses how sparse grids, which have been successfully used in numerical analysis to tackle high dimensional problems can be used to solve regression problems. We will especially focus on algorithms that allow problem based adaptive grid manipulation for iterative improvement of the approximation quality. We will also present the implementation of the algorithms in the sparse grid framework SG++ and discuss numerical results achieved on different datasets.