FACT. Machine learning analysis

Imaging Atmospheric Cherenkov Telescopes like FACT (First G-APD Cherenkov Telescope) produce a continuous flow of data during observation. One major task of a monitoring system is to detect changes in the gamma-ray flux of a source, and to alert other experiments if some predefined limit is reached in order to possibly trigger multi wavelength observations. Thus analyzing the data with low latency is essential for understanding the acceleration mechanisms in bright gamma-ray sources like active galactic nuclei. In order to calculate the fluxes of an observed source, it is necessary to calculate the instrument response function (IRF) and effectively minimize background noise. This analysis relies heavily on the usage of machine learning methods to perform background suppression and energy estimation. We describe how multi-variate models are applied to FACT's data stream with low latency, show IRFs, present fluxes and compare results to an existing analysis which does not use machine learning.