[Statlist] Talk on Statistics Friday, May 22, 2009

[Cecilia Rey]

We are glad to announce the following talk
*******************************************

*Friday, May 22, 2009, 15.15 HG G 19.1
with Mohamed Hebiri, Université Paris VII*

*******************************************

Title:
The Transductive LASSO

Abstract:
We consider the linear regression problem, where the number p of 
covariates is possibly larger than
the number n of observations, un- der sparsity assumptions. In this 
talk, we propose a generalized
version of the LASSO by Tibshirani (1996), based on geometrical remarks 
about the LASSO provi-
ded by Alquier and Hebiri (2008) and that takes into acount the ob 
jective of the statistician. As a
special case, we consider the problem of estimating the regression 
vector in the transductive setting
as described by Vapnik (1998), in which the estimator construction is 
based on a new unlabelled
dataset of interest. From a theoretical point of view, we derive 
Sparsity Inequalities (SI) for our
estimator, i.e., bounds on the estimation error involving the sparsity 
of the parameter we try to
estimate. We also derive an algorithm to provide an approximated 
solution for our estimator which
is based on the LARS algorithm (Efron et al. 2004).

The abstract is to be found under the following link: 
http://stat.ethz.ch/talks/research_seminar

-- 
ETH Zürich
Seminar für Statistik
Cecilia Rey-Lutz, HG G10.3
Rämistrasse 101                    
CH-8092 Zurich		                      	
mail: rey at stat.math.ethz.ch    	  		
phone: +41 44 632 3438/fax: +41 44 632 1228