Speaker
Description
Accurate measurements of the cosmic-ray electron and positron spectrum in the GeV–TeV energy range are essential to probe the existence of nearby astrophysical or exotic sources. However, existing measurements show discrepancies, particularly at TeV energies, where systematic uncertainties become more pronounced.
Previous Fermi-LAT analyses relied on supervised Machine Learning techniques for electron–positron event selection. Although effective, these methods require training on Monte Carlo simulations, making them inherently model-dependent and potentially subject to significant systematic uncertainties or even biases.
In this work, we propose an alternative approach based on Unsupervised Learning, which identifies patterns directly in the data and enables a nearly model-independent event selection. The method is applied to Fermi-LAT data, demonstrating its potential for astroparticle physics applications. The extension of this approach to the analysis of lower level information, rather than reconstructed variables alone, is currently under investigation.