\documentclass[11pt]{article}
\setlength{\oddsidemargin}{0.0truein}
\setlength{\evensidemargin}{0.0truein}
\setlength{\textwidth}{6.5truein}
\setlength{\topmargin}{0.0truein}
\setlength{\textheight}{9.0truein} \setlength{\headsep}{0.0truein}
\setlength{\headheight}{0.0truein} \setlength{\topskip}{10.0pt}
\setlength{\parskip}{5mm}
\usepackage{url}
\begin{document}

\begin{center}
  \textbf{\textsc{STANFORD UNIVERSITY}}\\[5pt]
  \textbf{\textsc{DEPARTMENT OF STATISTICS}}\\[5pt]
  \Large{\textbf\textsc{{BERKELEY-STANFORD JOINT COLLOQUIUM}}}\\[5pt]
  {\textsc{In honor of Jerry Friedman's coming retirement
}}
\end{center}

\begin{center}
  From 3:15 p.m., Tuesday, April 25, 2006\\
  Packard 101 with a reception in Sequoia Hall\\
\end{center}

\noindent{\bf Schedule:}\\
\begin{tabular}{cll}
3:15-4:15 &&Lecture by Andreas Buja\\
4:15-4:30 &&Break\\
4:30-5:30 &&Lecture by Michael Jordan\\
5:45-7:30 &&Reception with food and speeches\\
\end{tabular}

\vspace{15pt}
\begin{center}
  \textsl{ \bf Andreas Buja   }\\
Statistics Department; The Wharton School
University of Pennsylvania
\end{center}

\begin{center}
  \textbf{ Five Ideas of Jerry Friedman   } 
\end{center}

\noindent
For the last thirty years, Jerry Friedman has been one of the most
creative minds in our profession.  He has contributed an astounding
number of ideas with vast impact.  A shared feature of many of these
ideas is their algorithmic character combined with unusual conceptual
insight.

\noindent
In this talk I will discuss some of Jerry's older work: PRIM-9, CART,
ACE, MARS, and EPP, and maybe one recent piece of work: the
interpretation of boosting as additive modeling.  Although much of
Jerry's work is seen as the pursuit of predictive performance, many of
his methods have another angle: they are also tools for interpreting
data.  This is an aspect that may have been lost somewhat under the
impact of the recent encounter with machine learning.  It may be
time to restore a balance between predictive performance and the
more qualitative aspects of data analysis, and Jerry's work is most
suitable for making this point.

\vspace{15pt}
\begin{center}
  \textsl{   \bf Michael Jordan }\\
UC Berkeley
\end{center}

\begin{center}
  \textbf{ A Berkeley-Stanford Statistical Smorgasbord   } 
\end{center}

\noindent
In this talk I will present recent work on a number of statistical
themes that have their origins at Berkeley or Stanford.  The first
line of work that I will discuss goes back to Blackwell, who uncovered
links between the risk based on 0-1 loss and the class of f-divergences.
We extend this programme by linking f-divergences to the broader class
of so-called surrogate loss functions---computationally-inspired upper
bounds on 0-1 loss that have become central in the machine learning
literature on classification.  I will present applications to distributed
detection and sufficient dimension reduction.  In a second line of work,
I discuss some of the issues that arise when Stein's and Efron's shrinkage
and empirical Bayes ideas are deployed in the nonparametric Bayesian
setting of Dirichlet process priors, a setting whose foundations were
laid by Freedman, Doksum, Ferguson and others.  The Chinese restaurant
process studied by Berkeley and Stanford probabilists (e.g., Aldous,
Diaconis, Dubins and Pitman) turns out to be particularly helpful here.
I discuss applications in bioinformatics to haplotype inference
across subpopulations and to admixture problems.  The overall theme is
computationally-scalable statistical algorithms that target significant
real-world data analysis problems---and here I acknowledge a general
intellectual debt to another great Bay Area statistician: Jerry Friedman.

\noindent
Joint work with Peter Bartlett, Jon McAuliffe, XuanLong Nguyen,
Martin Wainwright, Yee Whye Teh, David Blei, and Eric Xing



\end{document}