Keyword: GPU
Paper Title Other Keywords Page
MOPOTK038 BPM Analysis with Variational Autoencoders network, focusing, diagnostics, optics 543
 
  • C.C. Hall, J.P. Edelen, J.A. Einstein-Curtis, M.C. Kilpatrick
    RadiaSoft LLC, Boulder, Colorado, USA
 
  Funding: This material is based upon work supported by the U.S. Department of Energy, Office of Science, Office of High Energy Physics under Award Number DE-SC0021699.
In particle accelerators, beam position monitors (BPMs) are used extensively as a non-intercepting diagnostic. Significant information about beam dynamics can often be extracted from BPM measurements and used to actively tune the accelerator. However, common measurement tools, such as measurements of kicked beams, may become more difficult when very strong nonlinearities are present or when data is very noisy. In this work, we examine the use of variational autoencoders (VAEs) as a technique to extract measurements of the beam from simulated turn-by-turn BPM data. In particular, we show that VAEs may have the possibility to outperform other dimensionality reduction techniques that have historically been used to analyze such data. When the data collection period is limited, or the data is noisy, VAEs may offer significant advantages.
 
DOI • reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-MOPOTK038  
About • Received ※ 09 June 2022 — Revised ※ 13 June 2022 — Accepted ※ 15 June 2022 — Issue date ※ 10 July 2022
Cite • reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)  
 
WEPOMS036 Accelerating Linear Beam Dynamics Simulations for Machine Learning Applications simulation, space-charge, controls, experiment 2330
 
  • O. Stein, I.V. Agapov, A. Eichler, J. Kaiser
    DESY, Hamburg, Germany
 
  Machine learning has proven to be a powerful tool with many applications in the field of accelerator physics. Training machine learning models is a highly iterative process that requires large numbers of samples. However, beam time is often limited and many of the available simulation frameworks are not optimized for fast computation. As a result, training complex models can be infeasible. In this contribution, we introduce Cheetah, a linear beam dynamics framework optimized for fast computations. We show that Cheetah outperforms existing simulation codes in terms of speed and furthermore demonstrate the application of Cheetah to a reinforcement-learning problem as well as the successful transfer of the Cheetah-trained model to the real world. We anticipate that Cheetah will allow for faster development of more capable machine learning solutions in the field, one day enabling the development of autonomous accelerators.  
DOI • reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-WEPOMS036  
About • Received ※ 07 June 2022 — Revised ※ 14 June 2022 — Accepted ※ 01 July 2022 — Issue date ※ 01 July 2022
Cite • reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)  
 
WEPOMS043 UFO, a GPU Code Tailored Toward MBA Lattice Optimization lattice, electron, simulation, optics 2346
 
  • M. Carlà, M. Canals
    ALBA-CELLS Synchrotron, Cerdanyola del Vallès, Spain
 
  The complexity of multi-bend achromatic optics is such that computational tools performance has become a dominant factor in the design process a last generation synchrotron light source. To relieve the problem a new code (UFO) tailored toward performance was developed to assist the design of the ALBA-II optics. Two main strategies contribute to the performance of UFO: the execution flow follows a data parallel paradigm, well suited for GPU execution; the use of a just-in-time compiler allows to simplify the computation whenever the lattice allows for it. At the core of UFO lies a parallel tracking routine structured for parallel simulation of optics which differs in some parameters, such as magnet strength or alignment, but retains the same element order, reflecting the scenario found in optimization processes, or when dealing with magnetic or alignment errors. Such an approach allows to take advantage of GPUs which yield the best performance when running thousands of parallel threads. Moreover UFO is not limited to tracking. A few modules that rely on the same tracking routine allow for the fast computation of dynamic and momentum aperture, closed orbit and linear optics.  
DOI • reference for this paper ※ https://doi.org/10.18429/JACoW-IPAC2022-WEPOMS043  
About • Received ※ 07 June 2022 — Revised ※ 16 June 2022 — Accepted ※ 19 June 2022 — Issue date ※ 21 June 2022
Cite • reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)