Manipulating the harmonic mode-locked regimes inside a fiber cavity by a reinforcement learning algorithm

Harmonic mode-locked fiber lasers provide generation of the ultrashort pulse train with high repetition rates up to gigahertz scale. However, setting appropriate parameters for the laser cavity to reach a harmonic mode-locked regime is often a non-trivial task. Depending on the dynamic of adjustment of the cavity elements one may reach unstable, multipulsing or harmonic mode-locked regimes at the same end-point parameters. Here, we demonstrate the state-of-theart fiber mode-locked laser assisted with reinforcement Soft Actor-Critic algorithm that is capable of learning a dynamic strategy of adjusting cavity parameters to maximize the order of harmonic mode-locked regime. Control of the pumping power and nonlinear transmission function of the state-of-the-art single walled carbon nanotube saturable absorber allows reaching a stable harmonic mode-locked regime.