Partial data querying through racing algorithms

Abstract

This paper studies the problem of learning from instances characterized by imprecise features or imprecise class labels. Our work is in the line of active learning, since we consider that the precise value of some partial data can be queried to reduce the uncertainty in the learning process. Our work is based on the concept of racing algorithms in which several models are competing. The idea is to identify the query that will help the most to quickly decide the winning model in the competition. After discussing and formalizing the general ideas of our approach, we study the particular case of binary SVM and give the results of some preliminary experiments.