Date: Tuesday 09 Dec 2025 – 15:00 (Europe/London)
Speaker: Andrea Santamaria Garcia, Lecturer in Artificial Intelligence for Particle Accelerators at the University of Liverpool
Abstract
Machine learning (ML) is a key technology for advancing particle accelerators and should play a central role in their future design. ML methods provide fast predictions at lower computational cost than analytical or classical numerical approaches, capture nonlinear correlations in data, and adapt to changes in machine conditions. These capabilities enable robust online detection, prediction, optimisation, and control, while also supporting accelerator design by reducing the cost of numerical simulations and guiding parameter searches in high-dimensional spaces. Among ML applications, optimisation is particularly prominent, with Bayesian optimisation and reinforcement learning emerging as leading paradigms. In this seminar, I will focus on tuning and control tasks in particle accelerators using these methods and demonstrate their performance in real machines.

Biography
Dr Andrea Santamaria Garcia is a lecturer in Artificial Intelligence for particle accelerators at the University of Liverpool and a member of the Cockcroft Institute. She is an Accelerator Physicist by training. She did her doctoral studies at CERN, where she studied crab cavity failures at the High Luminosity LHC. Following her PhD, she was awarded a CERN Senior Fellowship and joined the operations group at CERN, where she carried out targeted measurement campaigns for the LHC Injectors Upgrade project. In 2020, Andrea joined the Karlsruhe Institute of Technology, where she created and led the AI4Accelerators team. During this time, she coordinated a project focused on controlling ultrafast nonlinear phenomena using online reinforcement learning on hardware. She was also part of several projects focused on the development and deployment of machine learning algorithms in a variety of particle accelerators for tuning and control, including the development of one of the first differentiable beam physics packages, Cheetah. Finally, Andrea co-founded the Reinforcement Learning for Autonomous Accelerators (RL4AA) international collaboration, which organises workshops annually on the topic of reinforcement learning.