\magnification=1200
\baselineskip=20pt
\nopagenumbers
\font\big=cmr12 scaled \magstep2

\centerline{\bf STANFORD UNIVERSITY}
\centerline{\bf DEPARTMENT OF STATISTICS}
\centerline{\big DEPARTMENTAL SEMINAR}
\bigskip

\baselineskip=12pt
\centerline{4:15 p.m., Tuesday, October 19, 1999}
\centerline{Sequoia Hall Rm. 200}
\centerline{(Cookies at 3:45 in 1st Floor Lounge)}

\bigskip
\baselineskip=15pt
\centerline{\sl Jerome H. Friedman*}
\centerline{\sl Stanford University}
\bigskip
\centerline{\bf GRADIENT BOOSTING and MULTIPLE ADDITIVE REGRESSION TREES}
\centerline{\bf }
\bigskip

Abstract
 
Multivariate function estimation is viewed from the perspective of
numerical optimization in function space, rather than parameter space. A
gradient-descent "boosting" paradigm is established for rapid
fitting of additive basis function expansions, based on any differentiable
loss criterion. Focus is on the particular case where each of the basis
functions is a (CART) regression tree. Specialized tools are presented for
interpreting and visualizing such multiple additive regression tree models.
The goal is a class of fast procedures for regression and classification
that are competitive in accuracy, while being interpretable, and highly
robust especially when mining less than clean data. Connections with the
boosting methods of Freund and Schapire, and Friedman, Hastie and
Tibshirani are discussed.

\bigskip

* with help from Trevor Hastie and Rob Tibshirani.

\bye