Speaker
Description
SLAC MeV-UED is part of LCLS scientific user facility and has enabled unprecedented opportunities in the studies of ultrafast structural dynamics in a variety of gas, liquid and solid-state systems. To remain at the scientific and technical forefront, continuing enhancements to the facility and operations are needed. In this talk, we will describe developments of intelligent scientific facility at SLAC MeV-UED using state-of-the-art AI/ML techniques. Multi-objective Bayesian active learning was demonstrated for speeding up online beam tunings and giving trade-offs between key beam properties of interest. Two-stage constrained Bayesian optimization was conducted for improving valid data efficiency and enable PF learning with minimal human inputs. Meanwhile, smart decision-making algorithms are being developed for fast data analysis, autonomous sample explorations and efficient temporal domain sampling. In all, these developments will enable autonomous facility operations, maximize scientific outputs and open new areas in ultrafast science.
Footnotes
[1] F. Ji et al., Nat. Comm. 15, 4726 (2024)
[2] F. Ji et al., Proc. NAPAC’25, MOP094, Sacramento, CA, USA, Aug. 2025
[3] F. Ji et al., Proc. NAPAC’25, THP094, Sacramento, CA, USA, Aug. 2025
Funding Agency
This work was supported by the U.S. Department of Energy Office of Science, Office of Basic Energy Sciences under Contract No. DE-AC02-76SF00515
