Meta-learning architectures: Collecting, organizing and exploiting meta-knowledge

Abstract

While a valid intellectual challenge in its own right, meta-learning finds its real raison d'être in the practical support it offers Data Mining practitioners [20]. Indeed, the whole point of understanding how to learn in any given situation is to go out in the real world and learn as much as possible, from any source of data we encounter! However, almost any type of raw data will initially be very hard to learn from, and about 80% of the effort in discovering useful patterns lies in the clever preprocessing of data [47]. Thus, for machine learning to become a tool we can instantly apply in any given situation, or at least to get proper guidance when applying it, we need to build extended meta-learning systems that encompass the entire knowledge discovery process, from raw data to finished models, and that keep learning, keep accumulating meta-knowledge, every time they are presented with new problems. © 2011 Springer-Verlag Berlin Heidelberg.