Uncertainty Quantification and Robustness of Particle Accelerator Machine-Learning Based Models
O.R. Convery, A. Hanuka
SLAC, Menlo Park, California, USA
Y. Gal, L. Smith
Oxford University Press (Oxford Electronic Publishing), Oxford, United Kingdom
Funding:This work was supported by the Department of Energy, Laboratory Directed Research and Development program at SLAC National Accelerator Laboratory, under contract DE-AC02-76SF00515. Virtual Diagnostic (VD) is a computational tool built using deep learning that can be used to predict diagnostic outputs*. VDs are especially useful in systems where measuring outputs is invasive, limited, costly or runs the risk of modifying the output. In experiments with large ramifications, it is important to quantify the uncertainty of each prediction. Given out-of-distribution inputs (e.g. using the same machine in a different operation mode), it is also necessary to understand how robust the VD model is and how well it generalizes on unfamiliar data. In this work**, we use various compositions of neural networks to explore and enhance prediction uncertainty and robustness on data sets gathered from SLAC National Laboratory. We aim to accurately and confidently predict the longitudinal phase space images of electron beams. The ability to make informed decisions under uncertainty and limited computational power is crucial for reliable deployment of scalable deep learning tools on safety-critical systems such as particle accelerators. * Hanuka, A., Emma, C., Maxwell, T. et al. Sci Rep 11, 2945 (2021). ** Convery, O., Smith, L., Gal, Y., & Hanuka, A. Arxiv 2105.04654 (2021).
Progress Towards Machine Learning-Based Real-Time Non-Destructive Prediction of the Longitudinal Phase Space Evolution in a Particle Accelerator
C. Emma, O.R. Convery, A.L. Edelen, M.J. Hogan
SLAC, Menlo Park, California, USA
We discuss progress towards the implementation of machine learning based real-time prediction of the Longitudinal Phase Space (LPS) evolution in a particle accelerator*. This work focuses specifically on the implementation of three separate longitudinal phase space virtual diagnostics at the FACET-II accelerator facility: one at the exit of the photoinjector, one after the second bunch compressor and one and the end of the linac before the experimental area. We present simulation results describing the simultaneous prediction of the LPS in a single-bunch and two-bunch mode of operation and discuss design choices for the machine learning architecture selected for the LPS prediction task**. Finally we discuss initial experimental deployment of the virtual diagnostics in regular accelerator operations. *C. Emma et al., Phys. Rev. Accel. Beams 21, 112802 2018 **C. Emma et al., IBIC 2018 THBO01
Right click on video for Picture-in-Picture mode or Full screen display.
