Dr. John Langford of Yahoo Research to Speak at CS Seminar Tomorrow

Publish Date: 05/11/2010

Title: The Foundations of Interactive Learning
Speaker: Dr. John Langford, Yahoo Research
Date: Wednesday, May 12, 2010.
Time: 11:15 A.M. - 12:15 P.M.
Location: 3100 Torgersen Hall

Abstract

It is very natural to wish to apply machine learning to the large amounts of data generated by user interactions, but the process turns out to be delicate. For example, if a news story snippet is shown to a user, and the user clicks on (and reads) it, this is event is fundamentally not equivalent to a multiclass label for several reasons. For example, the user might easily have been interested in other (unseen) stories as well. To deal with these issues, we propose using the contextual bandit setting, where on each round information is used to choose an action, and feedback about just this action is observed. This setting has the great virtue that it's tractable, with algorithms enjoying regret and sample complexity guarantees entirely comparable to what's possible in a more familiar supervised learning setting.

About:

John Langford is a computer scientist, working as a senior researcher at Yahoo Research. His work includes research in machine learning, game theory, steganography, and Captchas. He studied physics and computer science at the California Institute of Technology, earning a double bachelor's degree in 1997, and received his Ph.D. from Carnegie Mellon University in 2002. He was previously a Research Associate Professor at the Toyota Technological Institute in Chicago. He has worked in the past at IBM's Watson Research Center in Yorktown, NY under the Goldstine Fellowship. He is also the author of the popular machine learning weblog, hunch.net.