flashcurve: A machine-learning approach for the simple and fast generation of adaptive-binning light curves with Fermi-LAT data

TL;DR

This paper introduces a deep learning method for rapidly generating adaptive-binning gamma-ray light curves from Fermi-LAT data, improving speed and accuracy over traditional approaches.

Contribution

It presents a novel machine learning approach to efficiently produce adaptive light curves, outperforming standard methods in speed and accuracy.

Findings

Fast and accurate adaptive binning with deep learning

Applicable to multi-messenger astrophysics

Prototype for other astrophysical data types

Abstract

Gamma rays measured by the Fermi-LAT satellite tell us a lot about the processes taking place in high-energetic astrophysical objects. The fluxes coming from these objects are, however, extremely variable. Hence, gamma-ray light curves optimally use adaptive bin sizes in order to retrieve most information about the source dynamics and to combine gamma-ray observations in a multi-messenger perspective. However, standard adaptive binning approaches are slow, expensive and inaccurate in highly populated regions. Here, we present a novel, powerful, deep-learning-based approach to estimate the necessary time windows for adaptive binning light curves in Fermi-LAT data using raw photon data. The approach is shown to be fast and accurate. It can also be seen as a prototype to train machine-learning models for adaptive binning light curves for other astrophysical messengers.