ABC-MART: Recent Improvements in Boosting, Trees and Classification Algorithms

Jerry Friedman’s MART algorithm has had considerable influence in search applications and other classification tasks. We introduce a new category of classification algorithms: Adaptive Base Class Boost (ABC-Boost). The talk will present a particular implementation of ABC-Boost called ABC-MART, a combination of ABC-Boost with MART. Extensive experiments demonstrate surprising improvements. From a coding perspective, the change is really easy to implement (involving just a few lines of code).

Speaker Details

Ping Li is an assistant professor at Cornell. His Ph.D. is from Stanford in Statistics (2007). Ping has perhaps the record for the most Microsoft Internships: one as a dev in Visual Studio, followed by three in MSR (two with Ken Church and one with Chris Burges). During the most recent internship, he popularized the use of MART, a method that was subsequently used by the MSRAD team in their top performing entry in the Ad Predict Contest, as well as a number of other search applications. Ping Li is among the 15 recipients of the ONR young investigator award in 2009.

Date:
Speakers:
Ping Li
Affiliation:
Cornell