Deep reinforcement learning control of white-light continuum generation

White-light continuum (WLC) generation in bulk media finds numerous applications ultrafast optics and spectroscopy. Due to the complexity of underlying spatiotemporal dynamics, WLC optimization typically follows empirical procedures. Deep reinforcement learning (RL) is a branch machine dealing with control automated systems using deep neural networks. In this Letter, we demonstrate capability RL agent generate long-term-stable from medium without any previous knowledge system dynamics or functioning. This work demonstrates that can be exploited effectively complex nonlinear optical experiments.