Exploring some practical issues of SVM+: Is really privileged information that helps?

Abstract

Learning using privileged information (LUPI) is a machine learning paradigm which aims at improving classification by taking advantage of information that is only available at training time —not at test time. SVM+ is an SVM-based implementation of LUPI. Despite this paradigm has potential interest for many applications, both LUPI and SVM+ have been scarcely explored up to date. In this work we report our effort in reproducing some results in the SVM+ literature and explore some practical issues of SVM+. The main finding is that just using randomly generated features as privileged information may perform similarly to using sensible (i.e. meaningful a priori) privileged information, at least in some problems.