CBOL-Tuner: Classifier-pruned Bayesian optimization to explore temporally structured latent spaces for particle accelerator tuning

TL;DR

The paper tackles tuning of high-dimensional, stochastic beam dynamics in particle accelerators by introducing CBOL-Tuner, which combines a CVAE for latent representation ($z \in \mathbb{R}^8$) and an LSTM for temporal dynamics with a DNN predictor, all fed into a classifier-pruned Bayesian optimizer that uses a ResNet50 classifier to filter non-physical latent trajectories. The optimization minimizes the total beam loss $L_{bt} = w^T L_b$ via the Expected Improvement acquisition function, enabling efficient latent-space exploration and identification of multiple optimal RF configurations. Compared with Random Search, CBOL-Tuner achieves lower median and mean $L_{bt}$ and reduced variability, while noting occasional hallucinations in the latent space due to limited training data. The framework promises practical impact for real-time accelerator tuning and can be extended to multi-objective optimization and larger datasets, leveraging the latent-evolution modeling approach.

Abstract

Complex dynamical systems, such as particle accelerators, often require complicated and time-consuming tuning procedures for optimal performance. It may also be required that these procedures estimate the optimal system parameters, which govern the dynamics of a spatiotemporal beam -- this can be a high-dimensional optimization problem. To address this, we propose a Classifier-pruned Bayesian Optimization-based Latent space Tuner (CBOL-Tuner), a framework for efficient exploration within a temporally-structured latent space. The CBOL-Tuner integrates a convolutional variational autoencoder (CVAE) for latent space representation, a long short-term memory (LSTM) network for temporal dynamics, a dense neural network (DNN) for parameter estimation, and a classifier-pruned Bayesian optimizer (C-BO) to adaptively search and filter the latent space for optimal solutions. CBOL-Tuner demonstrates superior performance in identifying multiple optimal settings and outperforms alternative global optimization methods.

Figures (4)

• Figure 1: Architecture of the CBOL-Tuner for beam optimization in particle accelerators.
• Figure 2: Three out of 15 2d projections of 6d phase space of charged particle beam in the LANSCE linear accelerator. Accelerating modules - 1 to 4 are 201 MHz drift tube linac (DTL) and 5 to 48 are 805 MHz coupled cavity linac (CCL). The beam serves various scientific areas like isotope production facility (IPF), ultra-cold neutrons (UCN), proton radiography (PRAD), weapons neutron research (WNR), proton storage ring (PSR).
• Figure 3: Target vs parameter search space for Bayesian optimization and classifier-pruned BO. Classifier-pruning eliminates latent trajectories which doesn't qualify the selection threshold.
• Figure 4: Boxplot comparison of different latent search methods: Bayesian optimization vs random search.