Biomedical research is being revolutionized by new technologies, such as gene microarrays, coupled with new methods for machine learning and statistical inference. The mRNA counts contained in microarrays provide a global view of cellular activity by simultaneously recording the expression levels of thousands of genes. Nonetheless, from the point of view of classical statistical modeling and inference, there are severe technical problems due to the small number of observations, typically tens, relative to the large number of genes, typically thousands. Working with Lei Xu, Aik Choon Ten and Dr. Rai Winslow from the Center for Cardiovascular Bioinformatics and Modeling, Professor Donald Geman is pursuing several new approaches to this type of "small-sample learning", including simplifying assumptions about the statistical dependency structure among genes. For instance, using experimental data to .nd characteristic expression patterns among pairs of genes, it is now possible to differentiate between healthy (blue) and diseased (red) tissues much more reliably. Hopefully, this will pave the way for new, ef.cient diagnostic techniques, as well as lead to a deeper understanding of disease and methods of treatment.