[Statlist] Seminar ueber Statistik

Christina Kuenzli kuenz|| @end|ng |rom @t@t@m@th@ethz@ch
Wed Nov 16 15:22:58 CET 2005


                       ETH and University of Zurich 
                                 Proff. 
A.D. Barbour - P. Buehlmann - F. Hampel - H.R. Kuensch - S. van de Geer

**************************************************************************     
           We are pleased to announce the following talks
**************************************************************************

Friday, November 18, 2005, 15.15, LEO C15

       Testing against a high-dimensional alternative, with
       applications to microarray data

       Jelle Goeman, Medical Statistics, University of Leiden

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

Friday, November 24, 2005, 15.15, LEO C15

       Model Selection Challenge

       Isabelle Guyon, ClopiNet, Berkeley, CA, USA

A central problem in predictive modeling is to select the best model for 
a given task. Several models have proven to perform well in various 
contexts, including kernel methods, neural networks, and decision trees. 
Yet, none of them outperforms consistently the others across 
applications. Additionally, the selection of  hyperparamenters also 
appears to be critical.

We have been working on the design of a benchmark for model selection in 
the context of supervised learning. We have set up a challenge platform 
for a competition
(http://www.modelselect.inf.ethz.ch/index.php), which 
is attracting an increasing number of participants. The participants are 
competing on five large two-class classification problems from different 
application domains.

Our presentation will review the following questions:

- What is the problem of model selection and what is so challenging 
  about it?
- What experience have we gained from  the organization of past challenges?
- How can we benchmark model selection for supervised learning?
- What limitations are we facing?
- What do we expect from the challenge?

The challenge participants have access to a library of machine learning 
algorithms based on the Spider 
(http://www.kyb.tuebingen.mpg.de/bs/people/spider/) developed at the Max 
Planck Institute. We will introduce the audience to the Spider environment.

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

(LEO (Leonhardstrasse 27, 8006 Zurich) is close to the main building,
across the hill-side station gof the 'Polybahn') 
Overview maps of ETH : http://www.ethz.ch/search/orientation_en.asp
________________________________________________________
Christina Kuenzli            <kuenzli using stat.math.ethz.ch>
Seminar fuer Statistik      
Leonhardstr. 27,  LEO D11      phone: +41 (0)44 632 3438         
ETH-Zentrum,                   fax  : +41 (0)44 632 1228 
CH-8092 Zurich, Switzerland        http://stat.ethz.ch/~




More information about the Statlist mailing list