[2E6-GS-10n] AI application

6:45 PM - 7:00 PM JST(9:45 AM - 10:00 AM UTC)

[2E6-GS-10n-06] Latent Space Analysis with Generative Models for Flavor Structure in Theoretical Particle Physics

〇Haruto Kitagawa¹, Satsuki Nishimura¹, Hajime Otsuka¹ (1. Kyushu University)

Keywords:

Machine Learning,Diffusion Model,Auto Encoder,Particle Physics,Neutrino Physics

In the Standard Model of particle physics, the origin of flavor, including the hierarchical structure of fermion masses and mixings, remains a mystery. Although numerous flavor models based on assumed symmetries have been proposed to address this problem, a model-independent understanding is still lacking. In this paper, we apply machine learning techniques to analyze features of the flavor structure without relying on specific symmetries. First, we generate numerous Yukawa matrices that reproduce experimental values using diffusion models. Second, to investigate nontrivial relationships among the resulting physical quantities, we visualize latent low-dimensional representations using autoencoders. Within this latent space, we identify the region where data consistent with experimental observables are encoded, expressed as inequalities. By decoding this region, we derive constraints satisfied by the observed values. This data-driven approach based on generative models may offer new insights into elucidating the flavor structure of particles.