[Statlist] Talk on statistics

Cecilia Rey rey @end|ng |rom @t@t@m@th@ethz@ch
Mon Nov 3 11:09:51 CET 2008


Invitation to a talk in the series Seminar über Statistik

Friday, November 7,  2008,  15.15 - 17.00
in LEO C6, Leonhardstrasse 27, 8092 Zurich

'Sparse Exponential Weighting' by Alexandre Tsybakov, CREST and 
University of Paris 6

The quality of solving several statistical problems, such as
adaptive nonparametric estimation, aggregation of estimators and
estimation under the sparsity scenario can be assessed in terms of
sparsity oracle inequalities (SOI) for the prediction risk. One of
the challenges is to build estimators that attain the sharpest SOI
under minimal assumptions on the dictionary. Methods of estimation
adapted to the sparsity scenario are mainly of the two types. Some
of them, like the BIC, enjoy nice theoretical properties without any
assumption on the dictionary but are computationally infeasible
starting from relatively modest dimensions $p$. Others, like the
Lasso, the Dantzig selector or their modifications, can be easily
realized for very large $p$ but their theoretical performance is
conditioned by severe restrictions on the dictionary. We will focus
on Sparse Exponential Weighting, a new method of sparse recovery
realizing a compromise between theoretical properties and
computational efficiency. The theoretical performance of the method
in terms of SOI is comparable with that of the BIC and is even
better in some aspects. No assumption on the dictionary is required.
At the same time, the method is computationally feasible for
relatively large dimensions $p$. It is constructed using an
exponential weighting with suitably chosen priors, and its analysis
is based on the PAC-Bayesian ideas in statistical learning.


The abstract is to be found under the following link:  
http://stat.ethz.ch/talks/research_seminar/2008.


The next talk takes place already *on Monday, November 10, 2008 * with 
John Haslett, Department of Statistics, Trinity College, Dublin 
('Sampling the palaeoclimate of the past 15000 years using Bayesian 
models' )
*ETH Zürich, Zentrum - ML H 37.1
Sonnegstrasse 3, 8092 Zürich Statistik*

Listeners are welcome!

-- 
ETH Zürich
Cecilia Rey-Lutz	          rey using stat.math.ethz.ch
Seminar für Statistik
Leonhardstr. 27, LEO D11	  phone: +41 44 632 34 38
CH-8092 Zurich, Switzerland	  fax  : +42 44 6321228




More information about the Statlist mailing list