Automated laser plasma accelerators

Laser-plasma wakefield accelerators are often cited for their potential to provide compact MeV-GeV particle and photon beams, and nearly as often for their shot-to-shot variation due to the highly non-linear nature of the underlying physics. They are highly sensitive to many parameters, and so precise tuning and maintaining output stability is an ongoing challenge for the application of plasma accelerators. To address this issue, many research groups are striving to develop on-line control and analysis of their experiments, allowing the opportunity to incorporate active feedback. Machine learning techniques, such as Bayesian optimization, can then provide efficient search algorithms for optimizing the plasma accelerator, as well as providing statistical models of the multi-dimensional parameter space to provide physical insight. Here, we will discuss results from recent work utilizing machine learning within plasma accelerators and the data-driven future of these machines.