Marcin Wolter, IFJ PAN, Krakow

"Support Vector Machines, Neural Networks and Kernel Methods in the 
analysis of HEP data"


High energy physicists started to use multivariate techniques and have
had considerable success in using them. Typical areas of application are
background suppression (classification) and parameter estimation 
(regression) where a physical quantity is extracted from a set of directly measured
observables. The reason to apply statistical learning multivariate methods
is, in most cases, simply the lack of knowledge about the mathematical
dependence of the quantity of interest on the relevant measured variables.
Multivariate methods are taking into account correlations between 
variables, therefore the better performance can be reached.

Three most popular multivariate algorithms are presented: neural networks,
kernel algorithms based on the probability distributions approximated by
kernel functions and the quite new method: Support Vector Machines (SVM).
In SVM algorithm data are transformed to a new high dimensional space.
This allows the linear separation of data by a separating hyperplane
defined by a set of support vectors.

After introducing the selected multivariate algorithms our applications
to the identification of soft electrons and tau leptons in the ATLAS 
experiment will be presented.