Learning from Data with Geometry-Aware Sparse Grids
Kilian Michael Röhner · 2020
Employing sparse grids for data mining mitigates the curse of dimensionality, is applicable for mining large datasets, and offers an explainable approach to machine learning. In this thesis, we introduce a sparse grid-based incremental learning scheme for density estimation supporting computational efficiency and online model adaptions. On top, we introduce geometry-aware sparse grids to tackle image classification problems and implement the methods in the user-friendly SG++ data mining pipeline.