The Electron-Ion Collider (EIC) represents a flagship project for the future of the Office of Science, Nuclear
Physics (NP) program, providing an unprecedented opportunity to deploy artificial intelligence (AI) and
machine learning (ML) techniques in the design stage of large-scale experiments. Recent advances in AI
promise to boost the traditional design process, which relies on manual inspection over Geant4 simulations
that are computationally expensive and not amenable to automated optimization.

We propose to use deep neural networks (DNNs) to transform Geant4 simulations into differentiable
models, which will make gradient-based optimization possible. As a case example, we will optimize the
granularity of the future EIC forward calorimeter to improve its energy resolution. We develop a model for
an optimized calorimeter design for the EIC and a corresponding high-fidelity DNN-based fast simulation
and reconstruction algorithm. At the end of this proposal, we will deliver a framework that will be
applicable for design of any experiments that rely on Geant4 simulations, which constitutes the backbone
for detector designs in nuclear physics, medical physics, and space science.

Our team combines expertise in AI/ML techniques applied to collider experiments with domain knowledge
of research and development (R&D) for calorimeters and EIC experiments. This proposal will expand our
existing fruitful partnerships between NP researchers and computer scientists at LLNL to include other
researchers from the California EIC consortium, which is a leading group in EIC R&D for tracking and
calorimetry. We will leverage this position to transfer the results from this project to technical design
reports, which are expected to converge in a few years.

To help build the AI-workforce of the future, we plan to engage graduate and undergraduate students from
UC Riverside, which is one of the few R1 minority-serving institutions. The students will visit LBNL or LLNL
during the summer, where they will gain exposure to the AI work developed in this project as well as other
applications in nuclear physics.

As noted in the recent AI for Nuclear Physics workshop, AI-driven design is an emerging field for the EIC.
Our team is uniquely positioned to capitalize on the once-in-a-multi-decade opportunity that the EIC
provides today. By providing a specific but impactful application centered on calorimetry, we also aim to
set the stage for AI applications in other subsystems and corresponding reconstruction software. We seek
to help achieve the goal of shaping the EIC to become the first “AI-driven collider” with the first “AI-driven
experiments.''