The so-called curse of dimensionality is still a road-block for the numerical treatment of higher-dimensional problems. The classical sparse grid approach, mitigating the curse of dimensionality, has been extended to allow us to tackle high-dimensional problems; to this end, spatially adaptive refinement, modified ansatz functions, and efficient regularization techniques have been the main focus. To show the universality of the adaptive sparse grid approach, different real-world applications have been studied, the scope ranging from classification and regression in Data Mining to likelihood approximations in astrophysics. In settings where high-dimensional problems feature low effective dimensionalities, we are now able to tackle problems in more than 150 dimensions with competitive results. Furthermore, a general and flexible toolbox has been developed, allowing to use spatially adaptive sparse grids for various applications.
«The so-called curse of dimensionality is still a road-block for the numerical treatment of higher-dimensional problems. The classical sparse grid approach, mitigating the curse of dimensionality, has been extended to allow us to tackle high-dimensional problems; to this end, spatially adaptive refinement, modified ansatz functions, and efficient regularization techniques have been the main focus. To show the universality of the adaptive sparse grid approach, different real-world applications ha...
»