Jun won Lee's PhD Proposal

[Christophe Giraud-Carrier]

Jun won Lee recently (and successfully) proposed his PhD research
proposal at Brigham Young University.

His thesis statement is as follows:
---
By clustering classification learning algorithms based on their
behavior over a large number of learning tasks, we devise a more
effective metalearning system for algorithm selection, which exhibits
high accuracy and offers new insight into several learning algorithms.
---
Jun has been working on analyzing algorithm behavior locally based on
a number of diversity measures rather than globally based on accuracy
only. He has been using these local measures to cluster learning
algorithms, which highlights interesting patterns of similarity,
sometimes across traditional model classes. This will also allow
reduction of the complexity of the metalearning task, by allowing
mapping to cluster
s of similarly-behaved algorithms rather than individual algorithms.

If you have any thoughts or suggestions, please feel free to post them
or email Jun directly.
Thanks,
Christophe