Ponente
Descripción
Machine Learning (ML) has been successfully applied across various domains, including
medical image analysis, remote sensing, computer vision, and engineering. Recently, the
astroparticle physics community has started employing ML algorithms for a wide range of
tasks. Among the most challenging is event selection, which entails distinguishing among
different species of cosmic rays detected by space instruments.
Traditionally, most experiments rely on Boosted Decision Trees or the higher-performing
Neural Networks, which are supervised learning methods. While effective, these
approaches are model-dependent, being trained on Monte Carlo simulations, and may thus
be sensitive to important systematic uncertainties or biases.
In this talk, I present an alternative approach based on Unsupervised Learning techniques,
that exploits their potential in detecting patterns without any guidance, enabling a completely
model-independent analysis. As a case study, I demonstrate the feasibility of using this
approach to identify cosmic-ray electrons in data collected by the Fermi-LAT gamma-ray
telescope. This methodology is versatile and can be extended to other scientific
applications.
