Speaker
Description
Recent advances in high-dimensional Bayesian Optimization have opened the door to new tools for beam commissioning. At the CERN Low Energy Ion Ring (LEIR), several tuning challenges arise from the complex parameter space governing beam transfer and accumulation dynamics. In this paper we benchmark several state-of-the-art High Dimensional Bayesian Optimization methods to optimize the transfer from Linac3 to LEIR and maximize the accumulated beam inside the ring. We evaluate algorithms based on different strategies: trust region approaches (TuRBO), sparse axis-aligned subspace priors (SAASBO), nested embeddings for mixed spaces (Bounce), and length-scale-adapted priors in regular Bayesian Optimization. Our results demonstrate the relative strengths of each method in the context of particle accelerator optimization, where sample efficiency is critical, the objective function exhibits sparsity in relevant dimensions, and the parameter space contains both local and global structures. The benchmarking provides practical insights for selecting appropriate algorithms for beam commissioning tasks, considering factors such as convergence speed, computational overhead, and robustness to noisy observations.
| In which format do you inted to submit your paper? | LaTeX |
|---|