Machine learning driven cosmic-ray reconstruction with radio antennas at IceCube Observatory

Cosmic-ray initiated air showers provide critical insights into high-energy astrophysical processes, yet accurately reconstructing physics observables using radio antennas remains a computational challenge. The methods currently employed in reconstruction using radio antennas require the generation of extensive simulation libraries for the reconstruction of each detected air shower, limiting scalability. In this talk, I present a novel approach to physics-observables reconstruction at the IceCube Observatory, leveraging radio antennas and Graph Neural Network techniques. By focusing on the emission collected only at the radio antennas, we employ an iterative technique to improve the accuracy of key shower parameters, such as direction, energy, and Xmax. This approach not only enhances the precision of reconstructions using only radio antennas but also offers a pathway to integrate large-scale data. I will discuss the key steps of the pipeline, the machine learning models employed, and the potential implications for future cosmic-ray studies.