A0203
Title: Qualitative robustness of divide-and-conquer methods for large data sets
Authors: Andreas Christmann - University of Bayreuth (Germany) [presenting]
Abstract: The topic is at the intersection of machine learning for big data and robust statistics. Divide-and-conquer methods play an important role in machine learning and big data. In robust statistics, there are five main notions of robustness: qualitative robustness, sensitivity curve, influence function, maxbias, and breakdown point. The focus will be on the qualitative robustness of machine learning methods using a divide-and-conquer approach for the big data situation. Special cases are distributed learning and localized learning.
