Author Affiliations
1College of Advanced Interdisciplinary Studies, National University of Defense Technology, Changsha 410073, Hunan, China2Institute for Quantum Information and the State Key Laboratory of High Performance Computing, College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, Hunan, China3Institute of Quantum Information Science and Technology, College of Science, National University of Defense Technology, Changsha 410073, Hunan, Chinashow less
Fig. 1. Architecture of multi-layer perceptron network
[29]. (a) Network structure; (b) mathematics process of perceptron
Fig. 2. Simple recurrent neural network and its computational graph
Fig. 3. Architecture of autoencoder
Fig. 4. Partition of material parameter space based on dielectric constant and permeability
[5] Fig. 5. Applications of metamaterials in various fields. (a) Optical communication
[68]; (b) biomedicine
[70]; (c) sensitive detection
[69] Fig. 6. Common metamaterial design methods
Fig. 7. Application of MLP in metamaterial design. (a) Using MLP to design nanoparticle
[89]; (b) using MLP series to solve multi-solution problems
[90]; (c) optimization and improvement of MLP—SMTL model
[34]; (d) optimization and improvement of MLP—DEIFS algorithm
[91] Fig. 8. Other deep learning methods for metamaterial design. (a) Using CNN to design binary imaged metasurface devices
[92]; (b) using GAN to design metamaterial devices with complex topology
[93]; (c) using DDQN to dynamically design device material types and structure parameters
[94]; (d) using DEDGO to design a variety of low-dimensional heterostructures
[95]; (e) using RNN to design metamaterial devices
[18] Fig. 9. Basic model of intelligent optimization algorithm in optical system
Fig. 10. Intelligent mode-locked laser structures based on different search methods. (a) Experimental structure of self-optimizing mode-locked laser based on genetic algorithm
[112]; (b) experimental structure of intelligent mode-locked fiber laser based on HLA algorithm
[113] Fig. 11. Intelligent mode-locked laser structures based on different deep reinforcement learning algorithms. (a) Adaptive mode-locked laser model designed using DQN algorithm
[115]; (b) automatic mode-locked laser system based on DELAY algorithm
[24]; (c) mode-locked laser system based on reinforcement learning and spectral learning
[116]; (d) intelligent mode-locked laser architecture based on TD3 algorithm
[117] Fig. 12. Control structure of partially coherent optical systems. (a) Coherent beam combination system for 4-beam fiber amplifier based on SPGD algorithm
[119]; (b) two-stage phase control system based on deep learning model and SPGD algorithm
[120] Fig. 13. Coherent optical neuron system
[123] Year | Algorithm name | Mode-locked state search time |
---|
2015 | EA[118] | ~30 min | 2016 | GA[112] | ~30 min | 2017 | GA[102] | ~30 s | 2019 | HLA[113] | 0.22 s least recovery time | 3.1 s average recovery time | 2021 | DELAY[24] | 0.472 s least recovery time | 1.948 s average recovery time | 2022 | MDRL[116] | 0.2 s least recovery time | 0.69 s average recovery time |
|
Table 1. Comparison of different algorithms in mode-locking time