Fig. 1. The experimental setup of ultra-short pulse measurement based on TS-DFT and time-lens
[14] Fig. 2. DNN architecture and supervised/unsupervised learning processes
[44]. (a) DNN architecture; (b) supervised learning process; (c) unsupervised learning process
Fig. 3. The experimental setup of TPA
[49]. (a) Intensity autocorrelation; (b) interference autocorrelation; (c) the training curve
Fig. 4. Single frame measurement scheme for femtosecond pulses
[47]. (a) Residual network for single frame measurement of femtosecond pulse intensity; (b)‒(e) single-shot characterization of time-domain intensity of femtosecond pulses with pulse widths of 100 fs, 500 fs, 1 ps, and 1.5 ps
Fig. 5. Dense-1D-U-Net neural network
[48]. (a) The architecture of Dense-1D-U-Net neural network; (b) dense connection structure
Fig. 6. The schematic diagram of fiber laser with computer-controlled feedback
[58] Fig. 7. Real-time intelligent MLFL
[62].(a) The experimental setup; (b) state search results shown from left to right, FML, second-order HML, third-order HML,
QS, and
QML; (c) comparison of initial lock time, recovery time and number of regimes between recent AML studies and our study; (d) schematic diagram of human-like algorithm; (e) time consumption of FML states on initial lock (blue dotted line) and recovery (red dotted line) in ten consecutive experiments; (f) operation records for 15 consecutive days
Fig. 8. Real-time AML based on genetic algorithm
[63]. (a) The experimental setup; (b) The flow chart of modified genetic algorithm; (c) comparison of time-consuming performance between ARS and improved genetic algorithm
Fig. 9. Computer-controlled smart laser
[65]. (a) The experimental setup; (b) comparison between optimal spectra obtained after optimizations for merit functions with different
α Fig. 10. Real-time AML for high repetition rate lasers
[66]. (a) The experimental setup for the pre-stretch technique enabled 48 MHz AML; (b) changes in the dispersion and CTR of lasers with different net dispersion at a fixed sampling rate of 500 MSa/s; (c) relationships between the dispersion, CTR, and different bandwidths at a fixed sampling rate of 500 MSa/s; (d) comparison of normalized waveforms at a fixed sampling rate and bandwidth; (e) relationships between the dispersion, CTR, and different sampling rates at a fixed bandwidth of 100 MHz
Fig. 11. The structure of low-latency deep reinforcement learning algorithm based on DDPG strategy
[73] Fig. 12. Spectral sequence feedback control method
[74]. (a) Spectral sequence feedback control model; (b) performance comparison of different algorithms
Fig. 13. The soliton searching process of spectrotemporal domain-informed deep learning
[75]. (a) Digital control system; (b) FCNN; (c) matching optimization process between experimental results and target soliton state; (d) experimental setup
Fig. 14. Breathing soliton MLFL device based on evolutionary algorithm
[76]. (a) Breathing soliton MLFL device; (b) the logic diagram of the evolutionary algorithm; (c) schematic diagram of roulette method; (d) schematic diagram of the breathing soliton RF signal
Fig. 15. Spatial-temporal mode-locked fiber laser based on wavefront shaping and genetic algorithm
[77]. (a) Device diagram; (b) mode profile after genetic optimization; (c) intensity evolution of the targeted area selected for genetic optimization; (d) wavelength tuning; (e) generation of multiple pulses
Fig. 16. Soliton molecular MLFL based on NPR
[78]. (a) Device diagram; (b) relationship between the soliton molecular spacing and spectral filtering parameter
β Fig. 17. Soliton molecular searching based on GA
[79]. (a) The laser setup; (b) spectral shaping based on SLM
Fig. 18. Soliton molecule phase control
[80]. (a) Schematic diagram; (b) the demonstration of phase coding experiment
Fig. 19. Real-time intelligent control scheme for dual comb based on a single resonant cavity
[81]. (a) Intelligent single-cavity double-comb; (b) the schematic diagram of MAIS; (c) time-consuming performance test; (d) stability test
Fig. 20. Real-time comprehensive control over soliton molecules
[82]. (a) The experimental setup for real-time comprehensive control over soliton molecules; (b) the spectra of soliton molecules (left) with inter-soliton separations ranging from 2 ps to 58 ps, accompanied by the corresponding autocorrelation traces (right); (c) the spectra of soliton molecules (left) with inter-soliton separations ranging from 3.0 ps to 3.7 ps with the interval of merely 0.1 ps, along with the corresponding autocorrelation traces (right); (d) comparison of target separations and searched separations (top), errors between target and searched separations (bottom)
Fig. 21. The flowchart of the evolutionary algorithm, which optimizes dispersion and driving parameters according to a given comb spectral contour
[84]