Speaker
Description
A(i)DAPT is a program aiming to utilize AI techniques, in particular
generative modeling, to support Nuclear and High Energy Physics
experiments. Its purpose is to extract physics directly from data in the
most complete manner possible. Generative models such GANs and
Normalizing Flows are employed to capture the full correlations between
particles in the final state of nuclear reactions. This many-fold
program will allow us to to achieve various goals including accurately
fitting data in a multidimensional space and unfolding detector effects
to minimize their impact on the relevant physics. Moreover, it will
enable us to store a large amount of realistic-like data in an extremely
compact format and to extract reaction amplitudes in an alternative way.
We aim at incorporating universality of scattering amplitudes, training
networks with different kinematics of the same final state or different
final states to recover the underlying physics. As of today, we've
conducted a positive closure test on inclusive electron scattering,
demonstrating that generative models are able to reproduce $2\pi$
photoproduction data. We also showed that GANs are a viable tool to
unfold detector smearing, ensuring the preservation of initial correlations.
