Yahoo! Research at Neural Information Processing Systems Conference (NIPS 2007)

Scenic Vancouver, B.C. and the Whistler Resort served as the locations for NIPS 2007, held from December 3 to 8. Yahoo! Research had a substantial presence, with Yahoo! researchers making a number of presentations. Among them, machine learning researcher John Langford presented a paper on learning for the ad display problem. When displaying an ad from a set of ads, the ad chosen for display is either clicked or not, but there's no feedback about other possible ads. John’s paper presented a technique that gives a provably sound algorithm for learning to choose which ads to display from user feedback.

At his workshop, Malcolm Slaney gave a talk on Music and Cognition titled “Hallucinations in Auditory Perception.” The goal of the workshop was to understand how people perceive sound. Music is an ideal sandbox for understanding human auditory perception because the difficulty of the signal can range from simple tones to full orchestral pieces. His workshop addressed several different aspects of machine perception.

Also reported were several advances in the theory of reinforcement learning, as well as important progress on large-scale applications. Alexander Strehl presented a poster titled "Online Linear Regression and Its Application to Model-Based Reinforcement Learning” (jointly done with Michael Littman, of Rutgers University) – a theoretical development of a new technique for solving certain reinforcement learning problems.

Olivier Chappelle co-organized a workshop for those interested in both machine learning and Web search. Andrei Broder, Research Fellow and Vice President, was an invited speaker at the workshop, which brought together machine learning and Web search researchers to discuss fundamental issues in Web search, from relevance ranking and Web spam detection to online advertising.