Speaker
Description
Modern experiments in fundamental physics increasingly rely on large-scale simulations and machine-learning techniques to extract rare signals from complex datasets. While these approaches have become indispensable, the computational cost of producing sufficiently large Monte Carlo datasets is rapidly becoming one of the major challenges for future precision measurements.
In this contribution, I present a Generative Adversarial Network (GAN)-based framework developed within the ALICE experiment at CERN for reconstructed-level data augmentation. The method generates high-fidelity synthetic signal samples from fully simulated events, enabling efficient augmentation of limited Monte Carlo statistics while preserving the multidimensional correlations required for physics analyses. The generated samples are validated through feature-space comparisons, correlation studies, and machine-learning-based compatibility tests.
Although the current application focuses on rare heavy-flavour hadron reconstruction in ALICE, the methodology is broadly applicable to computationally demanding problems across fundamental physics where simulations constitute a significant bottleneck. This work illustrates how generative AI can provide scalable solutions for future large-scale scientific experiments and highlights the growing role of modern machine-learning techniques in data-intensive physics research.