\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, November 18, 2003}
\centerline{Sequoia Hall Room 200}
\centerline{(Cookies at 3:45 in 1st Floor Lounge)}

\bigskip
\baselineskip=15pt
\centerline{\sl Joshua Sweetkind-Singer}
\centerline{\sl Stanford University}
\bigskip
\centerline{\bf Log-Penalized Linear Regression}
\bigskip

Abstract:

This talk discusses a form of linear regression that employs a log-like 
regularization penalty and compares this linear regression method 
against more common regularization penalties like the ridge and lasso 
penalties.  The penalty is motivated by appealing to MDL principles and 
algorithmic complexity theory.  It turns out that the log-penalty 
yields sparse solutions and has much in common with the lasso.  In 
fact, the log-penalized solutions can be efficiently approximated using 
an iterative weighted-lasso technique that exploits the computational 
efficiency of the LARS method for computing lasso solutions.  
Experimental results indicate that the log-penalized method yields 
better solutions than lasso when the true underlying model is extremely 
sparse.


\bye