Intelligent Mode-Locking Enabled by Real-Time Reinforcement Learning

A real-time embedded approach to realize intelligent mode-locked fiber lasers based on reinforcement learning is proposed. The reinforcement learning is deployed in a Field programmable gate array (FPGA) to complete the inference process and real-time control. Benefited from the dedicated quantization and deployment strategy, the FPGA finishes a single forward inference in 6.4 us, which is about 32-times faster than an Nvidia RTX-3050 GPU, and the proposed solution manifests ultrahigh power efficiency.