Semi-Supervised Learning

When a classifier is deployed in a particular environment, its performance could often be improved if it were adapted to that environment. We study how to adapt the initial classifier with a small set of labeled data form the new environment. We proposed a novel adaptive-training algorithm, which uses three additional classifiers to improve the performance of the initial. This is a join work with Božidara Cvetković, Mitja Luštrek and Matjaž Gams.