Berkeley Lab

Machine Learning, Artificial Intelligence, and Particle Accelerators

Recent advances in artificial intelligence (AI) and machine learning (ML) have produced powerful techniques that can be applied across many scientific domains. ATAP is working to apply AI and ML in the design and operation of particle accelerators.

Concentric colorful circles show beam spot size

Profile of the electron beam at the ALS, represented as pixels measured by a charged-coupled device (CCD) sensor. Demanding experiments require that the light-beam size be stable on time scales ranging from less than seconds to hours to ensure reliable data.

For instance, machine-learning based algorithms can help tune the parameters of an accelerator in real time, during its operation, in order to maximize performance. In particular, neural networks were recently used at the Advanced Light Source to improve beam-size stability by an order of magnitude. Similar techniques are being investigated and applied across various ATAP programs, including the Berkeley Lab Laser Accelerator Center (BELLA), the HiRES beamline for ultrafast electron diffraction, and ATAP’s laser development program.

Particle accelerators can exploit ML classification and anomaly detection algorithms, with the aim of preventing accelerator damage or beam loss in the case of abnormal operation. For example, ATAP recently applied ML techniques to the early detection and classification of quench precursors in superconducting magnets.

AI and ML can also be instrumental in the design of future accelerators. Accelerator design generally requires the optimization of a large number of coupled accelerator parameters through advanced numerical simulations. In this context, ML and AI methods can help identify the most promising combinations of parameters in advance, and thereby reduce the total number of simulations to be performed.

To Learn More…

●  “Machine Learning Paves Way for Smarter Particle Accelerators,” a news release by Will Ferguson of Berkeley Lab Strategic Communications.
●  “Machine Learning Enhances Light-Beam Performance at the Advanced Light Source,” a news release by Glenn Roberts, Jr., of Berkeley Lab Strategic Communications.
●  S.C. Leemann, S. Liu, A. Hexemer, M.A. Marcus, C.N. Melton, H. Nishimura, and C. Sun, “Demonstration of Machine Learning-Based Model-Independent Stabilization of Source Properties in Synchrotron Light Sources”, Phys. Rev. Lett. 123 (6 November 2019), 194801,
See also “Noisy Synchrotron? Machine Learning Has the Answer.” the Synopsis for this article by Senior Editor Katherine Wright in the magazine Physics
●  Yuping Lu, Changchun Sun, and Simon C. Leemann, “Improving Multi-objective Genetic Algorithm for Lattice Optimization with Machine Learning,” ALS User Meeting, 2020
●  Dan Wang, Qiang Du, Tong Zhou, Bashir Mohammed, Mariam Kiran, Derun Li, and Russell Wilcox, “Artificial neural networks applied to stabilization of 81-beam coherent combining,” in Proceedings of Optical Society of America Laser Congress 2020, paper ATu4A.6,
●  Qiang Du, Dan Wang, Tong Zhou, Derun Li, and Russell Wilcox, “Characterization and control for 81-beam diffractive coherent combining,” in Proceedings of Optical Society of America Laser Congress 2020, paper ATu4A.5,
●  Rémi Lehe, “Controlling Hundreds of GPU-Powered Plasma-Physics Simulations with Machine Learning Algorithms,” presentation at 2017 GPU Technology Conference (San Jose, CA, 8-11 May 2017).
●  Y. Lu, S. C. Leeman, C. Sun, M. P. Ehrlichman, H. Nishimura, M. Venturini, and T. Hellert. “Demonstration of machine learning-enhanced multi-objective optimization of ultrahigh-brightness lattices for 4th-generation synchrotron light sources.” Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 1050 (2023),


Point of contact for AI/ML topics in ATAP: Rémi Lehe,