Flexible Spectral-Normalized Neural Gaussian Process for Dynamic Aperture Prediction
Abstract
We address the challenge of scalable uncertainty quantification in large-scale scientific applications, where complex state-of-the-art machine learning methods are often computationally infeasible. Our primary contribution is a simple yet effective empirical Bayes method for automatically tuning the hyperparameters of a flexible, heteroscedastic Spectral-normalized Neural Gaussian Process. This approach retains the expressiveness and uncertainty-awareness of semi-Bayesian neural models while significantly reducing the computational burden by integrating hyperparameter learning directly into the training loop. We demonstrate the practical impact of our method on the task of estimating the dynamic aperture in circular particle accelerators, a fundamental problem in high-energy physics colliders and storage rings, using simulation data from the case of the Large Hadron Collider at CERN. Traditional approaches to DA estimation require extensive particle-tracking simulations, which are prohibitively time-consuming and resource-intensive. Our results show that the proposed method achieves competitive predictive performance and well-calibrated uncertainty estimates at much lower computational cost than state-of-the-art approaches. We stress that, beyond this application, the proposed empirical Bayes framework offers a general solution for training heteroscedastic neural models in situations where manual hyperparameter tuning is impractical. Accordingly, we anticipate that this framework can be applied to other domains that encounter comparable computational limitations.
Related
- Flexible Proposal Distributions for Gravitational-Wave Population Inference with Gaussian Mixture Models
- Spectral shaping of Gaussian white noise for synthetic axion signal generation in microwave cavity haloscopes
- Spectral contraction from Gaussian beams to Helmholtz waves
- The Global Geometry of the Gaussian Bures Manifold: Admissible Domain, Spectral Boundary, and Asymptotic Classicality
- Magnetic crystal generation with site-resolved magnetic moments via Gaussian-mixture retargeting and finite-budget yield prediction
BibTeX
@misc{elbachir2026flexible,
title = {{Flexible Spectral-Normalized Neural Gaussian Process for Dynamic Aperture Prediction}},
author = {Yousra El-Bachir and Frederik Van der Veken and Davide di Croce and Carlo Emilio Montanari and Massimo Giovannozzi and Ekaterina Krymova and Tatiana Pieloni},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.08620},
eprint = {2609.08620},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.08620},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Flexible Spectral-Normalized Neural Gaussian Process for Dynamic Aperture Prediction
AU - Yousra El-Bachir
AU - Frederik Van der Veken
AU - Davide di Croce
AU - Carlo Emilio Montanari
AU - Massimo Giovannozzi
AU - Ekaterina Krymova
AU - Tatiana Pieloni
PY - 2026
DA - 2026/09/08
PB - arXiv
DO - 10.48550/arxiv.2609.08620
UR - https://arxiv.org/abs/2609.08620
AB - We address the challenge of scalable uncertainty quantification in large-scale scientific applications, where complex state-of-the-art machine learning methods are often computationally infeasible. Our primary contribution is a simple yet effective empirical Bayes method for automatically tuning the hyperparameters of a flexible, heteroscedastic Spectral-normalized Neural Gaussian Process. This approach retains the expressiveness and uncertainty-awareness of semi-Bayesian neural models while significantly reducing the computational burden by integrating hyperparameter learning directly into the training loop. We demonstrate the practical impact of our method on the task of estimating the dynamic aperture in circular particle accelerators, a fundamental problem in high-energy physics colliders and storage rings, using simulation data from the case of the Large Hadron Collider at CERN. Traditional approaches to DA estimation require extensive particle-tracking simulations, which are prohibitively time-consuming and resource-intensive. Our results show that the proposed method achieves competitive predictive performance and well-calibrated uncertainty estimates at much lower computational cost than state-of-the-art approaches. We stress that, beyond this application, the proposed empirical Bayes framework offers a general solution for training heteroscedastic neural models in situations where manual hyperparameter tuning is impractical. Accordingly, we anticipate that this framework can be applied to other domains that encounter comparable computational limitations.
ER -