February 21
David Page
UW. Madison Department of Biostatistics and Medical Informatics, Department of Computer Science, and Comprehensive Cancer Center 
Supervised Learning for Gene Expression Microarray Data: Comparative Experiments with Multiple Myeloma

This talk compares leading supervised learning algorithms
on a new gene expression microarray data set for cancer with over
100 samples.  The paper provides evidence for several important
lessons, including (1) Bayes nets and simple ensembles perform at
least as well as other approaches but arguably provide more direct
insight and (2) looking for consistent differences in expression
may be more important than looking for large differences. The types
of supervised learning algorithms employed are decision trees,
boosted decision trees, support vector machines, unweighted voting
between decision stumps, and Bayes nets for classification, which
may be viewed as sophisticated weighted voting.  The data set is
based on Affymetrix "gene chip" (TM) technology and comes from a
study of multiple myeloma, a presently-incurable blood cancer.

This is joint work with Fenghuang Zhan, James Cussens, Mike Waddell,
Jo Hardin, Bart Barlogie, and John Shaughnessy.