Abstract

Online Ensemble Learning in Changing Environments

Lenadro Minku

School of Computer Science, University of Birmingham

Machine Learning is a branch of Artificial Intelligence which involves the study and development of computational systems capable of improving their performance with experience and of acquiring knowledge on their own. Most learning machines operate in offline mode, i.e., they first have learn a particular task and only afterwards they can be used to perform this task. No further learning can be done after the learning phase is finished. However, most real world problems change with time and the learning machines should be able to adapt to the changes. For example, in an information filtering system to predict what articles a user would like to read, the user may change their preferences with time. Differently from offline learning machines, online learning machines can learn during their entire existence. So, they can attempt to adapt to changes. We proposed a new approach to deal with these changes using ensembles of learning machines. The approach has shown to be accurate both in the presence and absence of drifts thanks to its novel feature of using knowledge acquired before changes in order to aid the learning after changes.