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

T: Fachverband Teilchenphysik

T 87: Datenanalyse

T 87.5: Vortrag

Donnerstag, 22. März 2018, 17:30–17:45, Z6 - SR 2.005

KM3NeT/ORCA data analysis using unsupervised Deep Learning — •Stefan Reck for the ANTARES-KM3NeT-Erlangen collaboration — Friedrich-Alexander-Universität Erlangen-Nürnberg, ECAP

KM3NeT/ORCA is a water-Cherenkov neutrino detector, currently under construction in the Mediterranean Sea at a depth of 2450 meters. Its main goal will be to determine the neutrino mass hierarchy by measuring the energy- and zenith angle dependency of the oscillation probabilities of atmospheric neutrinos after travelling through the Earth.

Deep Learning provides a promising method to analyse the signatures produced by the particles travelling through the detector. A common point of critique of the popular supervised Deep Learning techniques is their dependency on simulated data. If this data contains features that deviate from measured data, networks can become sensitive to them, and their performance on measurements will fall behind expectations. Ultimately, the network might fixate on effects only present in the simulations, or become unaware of properties of measured data.

This talk will cover an unsupervised learning approach with convolutional autoencoders, which tackles the problem of learning unwanted features by making it possible to train large parts of the network on measured data.