Fast Convergence Stochastic Parallel Gradient Descent Algorithm

Dongting Hu; Wen Shen; Wenchao Ma; Xinyu Liu; Zhouping Su; Huaxin Zhu; Xiumei Zhang; Lizhi Que; Zhuowei Zhu; Yixin Zhang; Guoqing Chen; Lifa Hu

doi:10.3788/LOP56.122201

Laser & Optoelectronics Progress, Volume. 56, Issue 12, 122201(2019)

Fast Convergence Stochastic Parallel Gradient Descent Algorithm

Dongting Hu^1,2, Wen Shen^1,2, Wenchao Ma^1,2, Xinyu Liu^1,2, Zhouping Su^1,2, Huaxin Zhu^1,2, Xiumei Zhang^1,2, Lizhi Que^1,2, Zhuowei Zhu^1,2, Yixin Zhang^1,2, Guoqing Chen^1,2, and Lifa Hu^1,2、*

Author Affiliations

¹ School of Science, Jiangnan University, Wuxi, Jiangsu 214122, China

² Jiangsu Provincial Research Center of Light Industrial Opto-Electronic Engineering and Technology, Wuxi, Jiangsu 214122, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(10)

Fig. 1. Block diagram of adaptive optics system without wavefront detection

Download full size

Fig. 2. Actuator distribution of 69-actuator deformable mirror

Download full size

Fig. 3. Flow chart of SPGD algorithm

Download full size

Fig. 4. Convergence for single random test and averaging after multiple tests under 800 iterations

Download full size

Fig. 5. Effects of perturbation amplitude δ and gain coefficient γ on residual wavefront and its optimal fitting curve

Download full size

Fig. 6. Optimal solution for wavefronts with different initial distortions. (a) Wavefronts with three initial distortion magnitudes; (b) distribution of optimal curve for wavefronts with different initial distortions

Download full size

Fig. 7. Convergence under different parameter combinations for wavefronts with different initial distortions. (a) 0.3312 rad;(b)0.8448 rad;(c)1.3180 rad

Download full size

Table 1. Convergence under different parameter combinations for wavefront with initial distortion RMS of 0.3312 rad
View table
Table 1. Convergence under different parameter combinations for wavefront with initial distortion RMS of 0.3312 rad
Algorithm SPGD SA
Disturbanceamplitude δ 1 1 1 N/A
Gaincoefficient γ -0.7 -1 -1.3 N/A
Wavefront RMS after800 iterations J₈₀₀ /rad 0.0258 0.0299 0.0361 0.0720
Convergenceefficiency η /% 92.21 90.97 89.10 78.26

Table 2. Convergence of different parameter combinations for wavefront with initial distortion RMS of 0.8448 rad
View table
Table 2. Convergence of different parameter combinations for wavefront with initial distortion RMS of 0.8448 rad
Algorithm SPGD SA
disturbanceamplitude δ 1 1 1 N/A
gaincoefficient γ -0.7 -1 -1.3 N/A
Wavefront RMS after800 iterations J₈₀₀ /rad 0.0708 0.0645 0.0736 0.1338
Convergenceefficiency η /% 91.62 92.37 91.29 84.16

Table 3. Convergence of different parameter combinations for wavefront with initial distortion RMS of 1.3180 rad
View table
Table 3. Convergence of different parameter combinations for wavefront with initial distortion RMS of 1.3180 rad
Algorithm SPGD SA
Disturbanceamplitude δ 1 1 1 N/A
Gaincoefficient γ -0.7 -1 -1.3 N/A
Wavefront RMS after800 iterations J₈₀₀ /rad 0.1098 0.0895 0.0877 0.1718
Convergenceefficiency η /% 91.65 93.17 93.32 86.97

Tools

Get Citation

Copy Citation Text

Dongting Hu, Wen Shen, Wenchao Ma, Xinyu Liu, Zhouping Su, Huaxin Zhu, Xiumei Zhang, Lizhi Que, Zhuowei Zhu, Yixin Zhang, Guoqing Chen, Lifa Hu. Fast Convergence Stochastic Parallel Gradient Descent Algorithm[J]. Laser & Optoelectronics Progress, 2019, 56(12): 122201

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Optical Design and Fabrication

Received: Nov. 29, 2018

Accepted: Jan. 9, 2019

Published Online: Jun. 13, 2019

The Author Email: Hu Lifa (hulifa@jiangnan.edu.cn)

DOI:10.3788/LOP56.122201

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Convergence under different parameter combinations for wavefront with initial distortion RMS of 0.3312 rad

Table 1. Convergence under different parameter combinations for wavefront with initial distortion RMS of 0.3312 rad

Table 2. Convergence of different parameter combinations for wavefront with initial distortion RMS of 0.8448 rad

Table 2. Convergence of different parameter combinations for wavefront with initial distortion RMS of 0.8448 rad

Table 3. Convergence of different parameter combinations for wavefront with initial distortion RMS of 1.3180 rad

Table 3. Convergence of different parameter combinations for wavefront with initial distortion RMS of 1.3180 rad

微信扫一扫：分享