Natural Sciences and Engineering Research Council of Canada
Symbol of the Government of Canada

Common menu bar links

Past Winner
1997 E.W.R. Steacie Memorial Fellowship

Robert Tibshirani

Biostatistics

University of Toronto


The scientist's world is a noisy place. Dr. Robert Tibshirani is trying to filter out some of that noise. The University of Toronto statistics professor uses statistics to find patterns amidst the random variations - the "noise" - of data. And find them he does. His research accomplishments are attracting international acclaim and recently earned him one of Canada's premier science and engineering prizes - a Steacie Fellowship awarded by the Natural Sciences and Engineering Research Council.

In one of his projects, Tibshirani is developing statistical techniques for a new computer-aided diagnosis system for mammograms. The system is being designed to provide a valuable second opinion to help radiologists in the difficult task of identifying tumours in the images. Tibshirani is also working with a Canadian company and computer scientists at the U of T to develop a new way for diabetics to monitor their blood sugar. The team hopes to replace diabetics' daily blood tests with a device that reads glucose levels by scanning the finger with infrared light.

"In both these applications there's a huge amount of data and it is extremely variable," explains Tibshirani. "We use statistics to make sense of the data, get rid of the noise, and deliver information that is of value to the user."

Part of this research involves using bootstrap methods, a simple and versatile technique for testing the accuracy of estimates, substituting computer power for theoretical analysis. Tibshirani literally wrote the book on the bootstrap with its inventor Bradley Efron.

In recent years, Tibshirani's research has brought him in contact with U of T computer scientist Geoffrey Hinton. Hinton is a pioneer in the field of neural networks, which are powerful modelling tools for complex systems used primarily by computer scientists.

"My collaboration with Geoff Hinton has been very refreshing," Tibshirani says. "Statisticians tend to be conservative about the applied problems they tackle, and the computational tools they use. As computer scientists, Geoff and his students have taught me to be more daring."

In fact, Tibshirani is now working with two colleagues at Stanford on a book that will pull statistics and neural networks together into a common framework which can be used by both statisticians and computer scientists.