Hamburg 2016 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 104: Gammaastronomie V

T 104.6: Vortrag

Donnerstag, 3. März 2016, 18:00–18:15, VMP9 SR 27

Active learning for Corsika — •Dominik Baack, Fabian Temme, Jens Buss, Max Nöthe, and Kai Brügge for the FACT collaboration — TU Dortmund, Dortmund, Deutschland

Modern Cosmic-Ray experiments need a huge amount of simulated data. In many cases, only a portion of the data is actually needed for following steps in the analysis chain,for example training of different machine learning algorithms. The other parts are thrown away by the trigger simulation of the experiment or so not increase the quality of following analysis steps.

In this talk, I present a new developed package for the air shower simulation software corsika. This extension includes different approaches to reduce the amount of unnecessary computation. One approach is a new internal particle stack implementation that allows to priorice the processing of special intermediate shower paricles and the removal of not needed shower particles.

The second approach is the possibility to sent various information of the initial particle and parameters of the status of the partial simulated event to an external application to approximate the information gain of the current simulator event. If the information gain is to low, the current event simulation gets terminated and all information get stored into a central database. For the Simulation - Server communication a simple network protocol has been developed.