Aachen 2019 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz

AKPIK 3: Machine-learning methods and computing in astroparticle physics

AKPIK 3.5: Talk

Wednesday, March 27, 2019, 16:40–16:50, H06

Search for new Source Populations with Autoencoding Neural Networks — •Simone Mender, Tobias Hoinka, and Kevin Schmidt — TU Dortmund

Active Galactic Nuclei (AGN) are astrophysical objects, whose emission range covers the entire electromagnetic spectrum. Based on the observations in different wavelengths, various source catalogs were published. In these catalogs, lots of unclassified sources are included. For example, the Fermi-LAT third source catalog (3FGL) contains 1010 unassociated sources.

In order to classify these sources, machine learning methods can be applied. With the use of supervised learning, it is possible to enlarge well-known source populations. To find new populations, unsupervised learning is a promising approach. Unknown source populations are expected to feature different characteristics compared to well-known populations. For outlier detection, an autoencoding neural network is a useful unsupervised machine learning technique. In this talk, we present our results of the application of autoencoding neural networks to gamma-ray and radio datasets.