Distinguished Lecture Abstracts

to main 25th Anniversary page

In this talk, David Salesin claims that the real market for computers lies in their vast potential as a communications medium. Already, millions of PowerPoint presentations are made each day, hundreds of thousands of documents are archived online, and billions of Web pages are searched. Yet, so far, computers are used largely just to emulate the appearance of existing, physical media, such as slide transparencies or 8½x11" sheets of paper. Drawing upon examples that range from computer-generated illustration and virtual cinematography to adaptive document layout and animated presentations, Salesin discusses some of the research challenges he sees in harnessing the power of the computer to create more powerful communications media than exist today.

David Salesin beginning his talk in the Lubrano conference room.

A typical medical environment is full of excellent (often expensive) hardware devices for gathering data about the state of a patient. Unfortunately, relatively little attention has been devoted to developing high quality software for integrating, analyzing, and presenting this information. This talk will briefly outline the problem, and then describe a novel hardware/software system designed to attack it. The talk will conclude with two applications: a machine-learning based system for early detection of epileptic seizures, and a system that combines an electronic stethoscope and software to screen for cardiac disease. Both applications have been tested on patients, and highly encouraging preliminary results will be reported.

This talk will focus on a general-purpose machine-learning method called boosting. The main idea of this method is to produce a very accurate classification rule by combining rough and moderately inaccurate "rules of thumb." While rooted in a theoretical framework of machine learning, boosting has been found to perform quite well empirically. In this talk, I will introduce the boosting algorithm AdaBoost, and explain the underlying theory of boosting, including our explanation of why boosting often does not suffer from overfitting, as well as some of the myriad other theoretical points of view that have been taken on this single algorithm. I also will describe some recent applications of boosting.