Sitzungen | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

FM: Fall Meeting

FM 82: Quantum & Information Science: Neural Networks, Machine Learning, and Artificial Intelligence III

FM 82.3: Talk

Donnerstag, 26. September 2019, 14:45–15:00, 3044

# aict-tools -- ML-based Event Reconstruction for Imaging Air Cherenkov Telescopes — •Maximilian Nöthe¹, Kai Arno Brügge¹, and Sabrina Einecke² — ¹Astroparticle Physics, TU Dortmund, Germany — ²Faculty of Sciences, University of Adelaide, Australia

Imaging Air Cherenkov Telescopes (IACTs) cover the highest energy ranges in the electromagnetic spectrum of astronomy.

These telescopes record the faint, nano-second scale flashes of Cherenkov radiation emitted by extensive air showers.

All IACTs face the same three reconstruction tasks, for each event, the primary particle's energy, direction and particle type have to be estimated. The particle type classification is necessary, as most extensive air showers are induced by charged cosmic rays.

Most commonly, IACTs record multiple time slices for each pixel in the camera for each shower, which is subsequently reduced to a few parameters describing each event.

The aict-tools use classical machine learning approaches as implemented by scikit-learn to reconstruct the gamma-ray properties from these image parameters.

Originally developed for the FACT Telescope, the library was extended to also work with data of the upcoming Cherenkov Telescope Array, e.g. the CHEC camera prototype.

The package provides executables to train, validate and apply models. It uses the yaml standard for defining configuration files and can store the resulting models in the pickle, pmml and onnx formats.