Imaging Atmospheric Cherenkov Telescopes (IACTs) enable precise ground-based observations of the gamma-ray sky by imaging the distribution of Cherenkov light emitted during the development of air showers.
Nowadays, many reconstruction algorithms rely on an elliptical high-level parameterization of these IACT images --- the Hillas parameterization --- and exploit their correlations.
To overcome the limitations of the elliptical modeling, besides sophisticated analytical or template-based models, the advent of deep learning allows for reconstruction techniques that showed first promising results.

By interpreting the detected images as a collection of triggered sensors that graphs can represent, we propose an algorithm based on graph networks for stereoscopic IACT image analyses.
For images cleaned from background noise, this allows for an efficient algorithm design that bypasses the challenge of sparse images that occur in deep learning approaches based on convolutional neural networks. We investigate graph network architectures to two different stereoscopic data sets, simulated for the H.E.S.S. experiment. The algorithm enables an excellent $\gamma$/hadron separation with improvements to classical machine learning. Further, we find that the algorithm offers promising prospects for stereoscopic reconstructions also for telescopes featuring different camera geometries.