Description:

In the Standard Model, the flavor sector contains the majority of free parameters corresponding to masses and mixings of the three generations of fermions. In the neutrino sector, these mixing parameters are being actively constrained by a number of current and upcoming oscillation experiments. Flavor models typically try to reduce the number of parameters by supposing that the fermions transform under a flavor symmetry, which would also explain the observed mixing. Constructing such a model is a detailed procedure, where one assigns charges under the flavor symmetry, diagonalizes mass matrices, and fits the mass and mixing parameters to observed quantities. Typically, theorists use these flavor models to explain other observed phenomena, such as dark matter or the origin of the matter/anti-matter asymmetry. In this talk, I will discuss a new technique to discover new models in the theory space. We have developed an automated model builder, AMBer, to construct new models of flavor. AMBer employs a reinforcement learning approach to develop new models by assigning charges, adding particles, fitting to observables, and learning how to build better models more quickly. We employ AMBer on the leptonic flavor sector and show that AMBer can discover many models in a number of different theory spaces, some of which have never been considered by humans. Thus, AMBer presents a new strategy to explore flavor models, removing a somewhat tedious process and allowing the physicist to focus on other important details in the design of useful flavor models. 

Max Fieg, Research Associate, Fermilab

Host: Adrian Thompson