B1146
Title: Robust multi-model subset selection
Authors: Anthony Christidis - University of British Columbia (Canada) [presenting]
Gabriela Cohen Freue - University of British Columbia (Canada)
Abstract: A method is proposed to learn an ensemble of sparse and robust models by leveraging recent developments in the robustness and ensemble literature. The degree to which the models are sparse, diverse and resistant to data contamination is driven directly by the data based on a cross-validation criterion. The finite-sample breakdown of the robust models in the ensembles is established, as well as the model itself is ensembled. A tailored computing algorithm is developed based on an extensive three-dimensional grid neighborhood search to generate solutions for any level of sparsity, diversity and robustness within the ensemble. The extensive numerical experiments on synthetic and real data sets demonstrate the competitive advantage of the method over the state-of-the-art high-dimensional robust methods.
