Machine Learning enhanced deterministic feedback controls in lasers and accelerators
by
Abstract: Lasers and accelerators are among the most complex scientific instruments, requiring control systems that are both fast and highly precise. As these systems continue to advance, traditional control methods can face challenges in meeting new performance demands. Machine Learning (ML) offers promising new tools to enhance feedback control by enabling adaptive, data-driven decision-making in real time.
In this talk, I will share recent efforts at LBNL to apply ML for improving control performance in laser systems. In particular, ML-based feedback has been demonstrated for shot-to-shot laser pointing stabilization on the BELLA PW beamline, helping to overcome conventional bandwidth limits. Similarly, ML-assisted coherent beam combining has achieved more than a tenfold improvement in response speed, illustrating the potential of ML for real-time Multi-Input Multi-Output (MIMO) control.
These studies highlight how ML can complement traditional approaches and contribute to advancing control capability in next-generation laser and accelerator systems.