This talk has two main parts. The first part will focus on the use of pair-wise meta-rules for algorithm ranking and selection. Such rules can provide interesting insights on their own, but they are also very valuable features for more sophisticated schemes like Random Forests. A hierarchical variant is able to address complexity issues when the number of algorithms to compare is substantial.
The second part of the talk will focus on meta-learning for data streams, which is a very active area of research currently. Stream algorithms need to be incremental, and be able to adapt to change in the distribution of the data. This poses new challenges for meta-learning.