Speaker
Description
We present a graph convolutional neural network (GCN) approach to analyzing the arrival directions of ultra-high-energy cosmic rays (UHECRs). By exploiting the sensitivity of GCNs to local spatial structure, the method enables the detection of small- and intermediate-scale anisotropies in UHECR arrival-direction distributions. Taking a set of active galactic nuclei (AGNs) as candidate sources, the model identifies patterns in UHECR sky maps and associates them with individual sources. We demonstrate the method by estimating the contributions of several nearby AGNs to the cosmic-ray flux observed by the Pierre Auger Observatory. We also identify individual events most likely to be associated with each candidate source, offering new opportunities to investigate cosmic-ray production and propagation in AGN environments.