Abstract
We establish a relaxation subgradient method (RSM) that includes parameter optimization using rank-two correction of metric matrices with a structure similar to that in quasi-Newtonian (QN) methods. The metric matrix transformation consists in suppressing orthogonal and amplifying collinear components of the minimum-length subgradient vector. The problem of constructing a metric matrix is stated as a problem of solving an involved system of inequalities. Solving such a system is based on a new learning algorithm. An estimate for its convergence rate is obtained depending on the parameters of the subgradient set. A new RSM has been developed and investigated on this basis. Computational experiments on complex large-dimension functions confirm the efficiency of the algorithm proposed.
Similar content being viewed by others
REFERENCES
N. Z. Shor, “Application of the gradient descent method for solving network transportation problems,” in Proc. Sci. Semin. Theor. Appl. Probl. Cybern. Oper. Res. (Nauchn. Sov. Kibern. Akad. Nauk UkrSSR, Kiev, 1962), no. 1, 9–17.
B. T. Polyak, “A general method for solving extremal problems,” Dokl. Akad. Nauk SSSR 174 (1), 33–36 (1967).
B. T. Polyak, Introduction to Optimization (Nauka, Moscow, 1983) [in Russian].
P. Wolfe, “Note on a method of conjugate subgradients for minimizing nondifferentiable functions,” Math. Program. 7 (1), 380–383 (1974).
E. G. Gol’shtein, A. S. Nemirovskii, and Yu. E. Nesterov, “The level method and its generalizations and applications,” Ekon. Mat. Metody 31 (3), 164–180 (1983).
Yu. E. Nesterov, “Universal gradient methods for convex optimization problems,” Math. Program. Ser. A 152, 381–404 (2015).
A. V. Gasnikov and Yu. E. Nesterov, Universal Method for Stochastic Composite Optimization (Cornell Univ., Ithaca, NY, 2016). Cornell Univ. Libr. e-Print Archive .
H. Ouyang and A. Gray, “Stochastic smoothing for nonsmooth minimizations: Accelerating SGD by exploiting structure,” in Proc. 29th Int. Conf. Mach. Learn. (Edinburgh, Scotland, June 26–July 1, 2012) (Omnipress, Madison, WI, 2012), pp. 33–40.
D. Boob, Q. Deng, and G. Lan, “Stochastic first-order methods for convex and nonconvex functional constrained optimization,” Math. Program. 2022 (in press). Available at https://doi.org/10.1007/s10107-021-01742-y . Accessed June 17, 2022.
G. Lan, First-Order and Stochastic Optimization Methods for Machine Learning (Springer, Cham, 2020).
S. Ghadimi and G. Lan, “Accelerated gradient methods for nonconvex nonlinear and stochastic programming,” Math. Program. 156 (1–2), 59–99 (2016).
C. Fang, C. J. Li, Z. Lin, and T. Zhang, “Spider: Near-optimal non-convex optimization via stochastic path-integrated differential estimator,” in Adv. Neural Inf. Process. Syst. 31, 32nd Annu. Conf. (Montreal, Canada, December 3–8, 2018) (Curran Assoc., Red Hook, NY, 2018), pp. 687–697.
A. S. Nemirovskii and D. B. Yudin, Complexity of Problems and Efficiency of Methods in Optimization (Nauka, Moscow, 1979) [in Russian].
N. Z. Shor, Minimization Methods for Nondifferentiable Functions and Applications (Nauk.Dumka, Kiev, 1979) [in Russian].
H. Cao, Y. Song, and K. Khan, “Convergence of subtangent-based relaxations of non-linear programs,” Processes 7 (4), 221 (2019).
B. T. Polyak, “Minimization of nonsmooth functionals,” Zh. Vychisl. Mat. Mat. Fiz. 9 (3), 509–521 (1969) [Comput. Math. Math. Phys. 9 (3), 14–29 (1969)].
V. N. Krutikov, N. S. Samoilenko, and V. V. Meshechkin, “On the properties of the method of minimization for convex functions with relaxation on the distance to extremum,” Avtom. Telemekh. (1), 126–137 (2019) [Autom. Remote Control 80 (1), 102–111 (2019)].
V. F. Dem’yanov and L. V. Vasil’ev, Non-differentiable Optimization (Nauka, Moscow, 1981) [in Russian].
C. Lemarechal, “An extension of Davidon methods to non-differentiable problems,” Math. Program. Study 3, 95–109 (1975).
V. N. Krutikov and T. V. Petrova, “Relaxation method of minimization with space extension in the subgradient direction,” Ekon. Mat. Metody 39 (1), 106–119 (2003).
V. N. Krutikov and T. A. Gorskaya, “A family of subgradient relaxation methods with rank \( 2 \) correction of metric matrices,” Ekon. Mat. Metody 45 (4), 37–80 (2009).
V. A. Skokov, “Note on minimization methods employing space stretching,” Kibern. Sist. Anal. (4), 115–117 (1974) [Cybern. Syst. Anal. 10 (4), 689–692 (1974)].
V. N. Krutikov and N. S. Samoilenko, “On the convergence rate of the subgradient method with metric variation and its applications in neural network approximation schemes,” Vestn. Tomsk. Gos. Univ. Ser. Mat. Mekh. (55), 22–37 (2018).
J. Nocedal and S. J. Wright, Numerical Optimization (Springer, New York, 2006).
M. Avriel, Nonlinear Programming: Analysis and Methods (Dover, Mineola, 2003).
E. A. Nurminskii and D. Tien, “Method of conjugate subgradients with constrained memory,” Avtom. Telemekh. (4), 67–80 (2014) [Autom. Remote Control 75 (4), 646–656 (2014)].
Ya. Z. Tsypkin, Basics of Theory of Learning Systems (Nauka, Moscow, 1970) [in Russian].
E. L. Zhukovskii and R. Sh. Liptser, “A recurrence method for computing the normal solutions of linear algebraic equations,” Zh. Vychisl. Mat. Mat. Fiz. 12 (4), 843–857 (1972) [Comput. Math. Math. Phys. 12 (4), 1–18 (1972)].
V. N. Krutikov, L. A. Kazakovtsev, and V. L. Kazakovtsev, “Non-smooth regularization in radial artificial neural networks,” IOP Conf. Ser. Mater. Sci. Eng. 450 (4), 042010 (2018).
V. N. Krutikov, L. A. Kazakovtsev, G. Sh. Shkaberina, and V. L. Kazakovtsev, “New method of training two-layer sigmoid neural networks using regularization,” IOP Conf. Ser. Mater. Sci. Eng. 537 (4), 042055 (2019).
R. J. Tibshirani, “Regression shrinkage and selection via the Lasso,” J. R. Stat. Soc. Ser. B 58, 267–288 (1996).
R. Frostig, R. Ge, S. M. Kakade, and A. Sidford, “Un-regularizing: Approximate proximal point and faster stochastic algorithms for empirical risk minimization,” Proc. Mach. Learn. Res. 37, 2540–2548 (2015).
Funding
This research was supported by the Ministry of Science and Higher Education of the Russian Federation, state contract no. FEFE–2020–0013. The research by the second author was supported by the Science Foundation of the Republic of Serbia, grant no. 7750185, and the Ministry of Education, Science, and Technological Development of the Republic of Serbia, contract no. 451–03–68/2020–14/200124.
Author information
Authors and Affiliations
Corresponding authors
Additional information
Translated by V. Potapchouck
Rights and permissions
About this article
Cite this article
Krutikov, V.N., Stanimirovi’c, P.S., Indenko, O.N. et al. Optimization of Subgradient Method Parameters Based on Rank-Two Correction of Metric Matrices. J. Appl. Ind. Math. 16, 427–439 (2022). https://doi.org/10.1134/S1990478922030073
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1134/S1990478922030073