An Introduction to Active Learning

David Cohn
Just Research

Place and Time

1305 Newell-Simon Hall
Refreshments 3:15 pm
Talk 3:30 pm

Abstract

Traditionally, most machine learning research has treated learning as a passive phenomenon, and has treated the learner as a passive recipient of data to be processed. This approach ignores the fact that, in many situations, the learner's most powerful tool is its ability to act, to gather data, and to influence the world it is trying to understand. Active learning is the study of how to use this ability effectively.

I will review the general framework of active learning and give illustrative examples from several applications. I will show how, with suitable abuse of statistics, one can compute the "optimal" way to do active learning on a variety of machine learning architectures. As time permits, I will also describe my recent work on "active learning on a fixed budget" and discuss preliminary results and open problems surrounding it.

Speaker Biography

He graduated Dartmouth College 1985 with an A.B. in Computer Science and Physics, then pursued (and eventually caught) his Ph.D. at the University of Washington. His thesis examined the large discrepancies between theoretical bounds on the performance of machine learning algorithms and the practical performance actually observed.

From 1992-1995, he worked worked as a postdoctoral researcher at the MIT Department of Brain and Cognitive Sciences. In 1995 he became a founding member of Harlequin, Inc.'s Adaptive Systems Group, where he applied machine learning to everything he could get his hands on, including adaptive scheduling, compiler optimization, hazardous waste remediation, memory management and document clustering.

He has been with Just Research since January of this year.

Speaker Appointments

For appointments, please contact the host, Andrew Moore (awm@ri.cmu.edu).