Improved SVRG for finite sum structure optimization with application to binary classification

Guangmei Shao; Wei Xue; Gaohang Yu; Xiao Zheng

doi:10.3934/jimo.2019052

Article Contents

2020, Volume 16, Issue 5: 2253-2266. Doi: 10.3934/jimo.2019052

This issue Previous Article Analytical modeling of laminated composite plates using Kirchhoff circuit and wave digital filters Next Article Differential equation method based on approximate augmented Lagrangian for nonlinear programming

Improved SVRG for finite sum structure optimization with application to binary classification

1.
School of Computer Science and Technology, Anhui University of Technology, Maanshan 243032, China
2.
School of Sciences, Hangzhou Dianzi University, Hangzhou 310018, China

^* Corresponding author: Wei Xue (E-mail: cswxue@ahut.edu.cn)
^* Corresponding author: Wei Xue (E-mail: cswxue@ahut.edu.cn)

Received: April 30, 2018

Revised: October 31, 2018

Early access: May 2019

Published: August 2020

Abstract Full Text(HTML) Figure(9) / Table(1) Related Papers Cited by

Abstract

This paper looks at a stochastic variance reduced gradient (SVRG) method for minimizing the sum of a finite number of smooth convex functions, which has been involved widely in the field of machine learning and data mining. Inspired by the excellent performance of two-point stepsize gradient method in batch learning, in this paper we present an improved SVRG algorithm, named stochastic two-point stepsize gradient method. Under some mild conditions, the proposed method achieves a linear convergence rate $ O(\rho^k) $ for smooth and strongly convex functions, where $ \rho\in(0.68, 1) $. Simulation experiments on several benchmark data sets are reported to demonstrate the performance of the proposed method.

Keywords:

Mathematics Subject Classification: Primary: 65K05; Secondary: 68W27.

Citation:

Full Text(HTML)

Figure 1. Objective loss and test classification accuracy versus epochs on data set "a9a" (best viewed in color)

Download: Full-size image PowerPoint slide

Figure 2. Objective loss and test classification accuracy versus epochs on data set "w8a"

Download: Full-size image PowerPoint slide

Figure 3. Evolutions of objective loss and test classification accuracy w.r.t. epochs for Eq. (14) on data set "a9a"

Download: Full-size image PowerPoint slide

Figure 4. Evolutions of objective loss and test classification accuracy w.r.t. epochs for Eq. (14) on data set "w8a"

Download: Full-size image PowerPoint slide

Figure 5. Evolutions of objective loss and test classification accuracy w.r.t. epochs for Eq. (14) on data set "a1a"

Download: Full-size image PowerPoint slide

Figure 6. Evolutions of objective loss and test classification accuracy w.r.t. epochs for Eq. (14) on data set "w1a"

Download: Full-size image PowerPoint slide

Figure 7. Evolutions of objective loss and test classification accuracy versus epochs on data set "a9a" with different $ \eta_0 $

Download: Full-size image PowerPoint slide

Figure 8. Evolutions of objective loss and test classification accuracy versus epochs on data set "w8a" with different $ \eta_0 $

Download: Full-size image PowerPoint slide

Figure 9. Comparison results between SVM and STSG on test classification accuracy on four test data sets

Download: Full-size image PowerPoint slide

Table 1. Details of data sets

Data set $\sharp$ of training samples $\sharp$ of test samples $\sharp$ of dimension

a1a 1605 30956 123
a9a 32561 16281 123
w1a 2477 47272 300
w8a 49749 14951 300

| Show Table

DownLoad: CSV

Related Papers

Cited by

References

[1]	J. Barzilai and J. M. Borwein, Two-point step size gradient methods, IMA J. Numer. Anal., 8 (1988), 141-148. doi: 10.1093/imanum/8.1.141.
[2]	S. Boyd and L. Vandenberghe, Convex Optimization, Cambridge University Press, New York, 2004. doi: 10.1017/CBO9780511804441.
[3]	Y. Chen, L. Qi and Q. Wang, Computing extreme eigenvalues of large scale hankel tensors, J. Sci. Comput., 68 (2016), 716-738. doi: 10.1007/s10915-015-0155-8.
[4]	W. Cheng and Y. Dai, Gradient-based method with active set strategy for l₁ optimization, Math. Comp., 87 (2018), 1283-1305. doi: 10.1090/mcom/3238.
[5]	Y. Dai and R. Fletcher, On the asymptotic behaviour of some new gradient methods, Math. Program., 103 (2005), 541-559. doi: 10.1007/s10107-004-0516-9.
[6]	Y. Dai and R. Fletcher, Projected Barzilai-Borwein methods for large-scale box-constrained quadratic programming, Numer. Math., 100 (2005), 21-47. doi: 10.1007/s00211-004-0569-y.
[7]	R. Johnson and T. Zhang, Accelerating stochastic gradient descent using predictive variance reduction, in Adv. Neural Inf. Process. Syst., (2013), 315–323.
[8]	H. Liu, Z. Liu and X. Dong, A new adaptive Barzilai and Borwein method for unconstrained optimization, Optim. Lett., 12 (2018), 845-873. doi: 10.1007/s11590-017-1150-9.
[9]	J. Mairal, Incremental majorization-minimization optimization with application to large-scale machine learning, Optim. Lett., 25 (2015), 829-855. doi: 10.1137/140957639.
[10]	E. Min, Y. Zhao, J. Long, C. Wu, K. Li and J. Yin, SVRG with adaptive epoch size, in Proc. IEEE Int. Joint Conf. Neural Netw., (2017), 2935–2942. doi: 10.1109/IJCNN.2017.7966219.
[11]	Y. Nesterov, Introductory Lectures on Convex Optimization: A Basic Course, Springer, Boston, 2004. doi: 10.1007/978-1-4419-8853-9.
[12]	A. Nitanda, Accelerated stochastic gradient descent for minimizing finite sums, in Proc. Int. Conf. Artif. Intell. Statist., (2016), 195–203.
[13]	A. Nitanda, Stochastic proximal gradient descent with acceleration techniques, in Adv. Neural Inf. Process. Syst., (2016), 1574–1582.
[14]	W. B. Powell, Approximate Dynamic Programming: Solving the Curses of Dimensionality, John Wiley and Sons, New Jersey, 2007. doi: 10.1002/9780470182963.
[15]	M. Raydan, On the Barzilai and Borwein choice of steplength for the gradient method, IMA J. Numer. Anal., 13 (1993), 321-326. doi: 10.1093/imanum/13.3.321.
[16]	H. Robbins and S. Monro, A stochastic approximation method, Ann. Math. Statist., 22 (1951), 400-407. doi: 10.1214/aoms/1177729586.
[17]	N. L. Roux, M. Schmidt and F. Bach, A stochastic gradient method with an exponential convergence rate for finite training sets, in Adv. Neural Inf. Process. Syst., (2012), 2663–2671.
[18]	Z. Shen, H. Qian, T. Zhou and T. Mu, Adaptive variance reducing for stochastic gradient descent, in Proc. Int. Joint Conf. Artif. Intell., (2016), 1990–1996.
[19]	C. Tan, S. Ma, Y. Dai and Y. Qian, Barzilai-Borwein step size for stochastic gradient descent, in Adv. Neural Inf. Process. Syst., (2016), 685–693.
[20]	Z. Wen, C. Yang, X. Liu and Y. Zhang, Trace-penalty minimization for large-scale eigenspace computation, J. Sci. Comput., 66 (2016), 1175-1203. doi: 10.1007/s10915-015-0061-0.
[21]	L. Xiao and T. Zhang, A proximal stochastic gradient method with progressive variance reduction, SIAM J. Optim., 24 (2014), 2057-2075. doi: 10.1137/140961791.
[22]	W. Xue, W. Zhang and J. Ren, Online leaming based on stochastic spectral gradient, Comput. Sci., 43 (2016), 47-51.
[23]	G. Yu and S. Niu, Multivariate spectral gradient projection method for nonlinear monotone equations with convex constraints, J. Ind. Manag. Optim., 9 (2013), 117-129. doi: 10.3934/jimo.2013.9.117.
[24]	B. Zhou, L. Gao and Y. Dai, Gradient methods with adaptive step-sizes, Comput. Optim. Appl., 35 (2006), 69-86. doi: 10.1007/s10589-006-6446-0.
[25]	R. Zhou, X. Shen and L. Niu, A fast algorithm for nonsmooth penalized clustering, Neurocomputing, 273 (2018), 583-592. doi: 10.1016/j.neucom.2017.08.048.