Skip to main content
Erschienen in: Programming and Computer Software 6/2023

01.12.2023

Gradient-Free Algorithms for Solving Stochastic Saddle Optimization Problems with the Polyak–Łojasiewicz Condition

verfasst von: S. I. Sadykov, A. V. Lobanov, A. M. Raigorodskii

Erschienen in: Programming and Computer Software | Ausgabe 6/2023

Einloggen

Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.

search-config
loading …

Abstract

This paper focuses on solving a subclass of stochastic nonconvex-nonconcave black box optimization problems with a saddle point that satisfy the Polyak–Łojasiewicz (PL) condition. To solve this problem, we provide the first (to our best knowledge) gradient-free algorithm. The proposed approach is based on applying a gradient approximation (kernel approximation) to an oracle-biased stochastic gradient descent algorithm. We present theoretical estimates that guarantee its global linear rate of convergence to the desired accuracy. The theoretical results are checked on a model example by comparison with an algorithm using Gaussian approximation.

Sie haben noch keine Lizenz? Dann Informieren Sie sich jetzt über unsere Produkte:

Springer Professional "Technik"

Online-Abonnement

Mit Springer Professional "Technik" erhalten Sie Zugriff auf:

  • über 67.000 Bücher
  • über 390 Zeitschriften

aus folgenden Fachgebieten:

  • Automobil + Motoren
  • Bauwesen + Immobilien
  • Business IT + Informatik
  • Elektrotechnik + Elektronik
  • Energie + Nachhaltigkeit
  • Maschinenbau + Werkstoffe




 

Jetzt Wissensvorsprung sichern!

Springer Professional "Wirtschaft+Technik"

Online-Abonnement

Mit Springer Professional "Wirtschaft+Technik" erhalten Sie Zugriff auf:

  • über 102.000 Bücher
  • über 537 Zeitschriften

aus folgenden Fachgebieten:

  • Automobil + Motoren
  • Bauwesen + Immobilien
  • Business IT + Informatik
  • Elektrotechnik + Elektronik
  • Energie + Nachhaltigkeit
  • Finance + Banking
  • Management + Führung
  • Marketing + Vertrieb
  • Maschinenbau + Werkstoffe
  • Versicherung + Risiko

Jetzt Wissensvorsprung sichern!

Springer Professional "Wirtschaft"

Online-Abonnement

Mit Springer Professional "Wirtschaft" erhalten Sie Zugriff auf:

  • über 67.000 Bücher
  • über 340 Zeitschriften

aus folgenden Fachgebieten:

  • Bauwesen + Immobilien
  • Business IT + Informatik
  • Finance + Banking
  • Management + Führung
  • Marketing + Vertrieb
  • Versicherung + Risiko




Jetzt Wissensvorsprung sichern!

Anhänge
Nur mit Berechtigung zugänglich
Literatur
1.
Zurück zum Zitat Goodfellow, I., Bengio, Y., and Courville, A., Deep Learning, MIT Press, 2016.MATH Goodfellow, I., Bengio, Y., and Courville, A., Deep Learning, MIT Press, 2016.MATH
2.
Zurück zum Zitat Dai, B., et al., SBEED: Convergent reinforcement learning with nonlinear function approximation, Proc. Int. Conf. Machine Learning, 2018, pp. 1125–1134. Dai, B., et al., SBEED: Convergent reinforcement learning with nonlinear function approximation, Proc. Int. Conf. Machine Learning, 2018, pp. 1125–1134.
3.
Zurück zum Zitat Namkoong, H. and Duchi, J.C., Variance-based regularization with convex objectives, Adv. Neural Inf. Process. Syst., 2017, vol. 30. Namkoong, H. and Duchi, J.C., Variance-based regularization with convex objectives, Adv. Neural Inf. Process. Syst., 2017, vol. 30.
4.
Zurück zum Zitat Xu, L., et al., Maximum margin clustering, Adv. Neural Inf. Process. Syst., 2004, vol. 17. Xu, L., et al., Maximum margin clustering, Adv. Neural Inf. Process. Syst., 2004, vol. 17.
5.
Zurück zum Zitat Sinha, A., et al., Certifying some distributional robustness with principled adversarial training, 2017. Sinha, A., et al., Certifying some distributional robustness with principled adversarial training, 2017.
6.
Zurück zum Zitat Audet, C. and Hare, W., Derivative-free and blackbox optimization, 2017. Audet, C. and Hare, W., Derivative-free and blackbox optimization, 2017.
7.
Zurück zum Zitat Rosenbrock, H.H., An automatic method for finding the greatest or least value of a function, Comput. J., 1960, vol. 3, no. 3, pp. 175–184.MathSciNetCrossRef Rosenbrock, H.H., An automatic method for finding the greatest or least value of a function, Comput. J., 1960, vol. 3, no. 3, pp. 175–184.MathSciNetCrossRef
8.
Zurück zum Zitat Gasnikov, A., et al., Randomized gradient-free methods in convex optimization, 2022. Gasnikov, A., et al., Randomized gradient-free methods in convex optimization, 2022.
9.
Zurück zum Zitat Lobanov, A., et al., Gradient-free federated learning methods with l 1 and l 2-randomization for non-smooth convex stochastic optimization problems, 2022. Lobanov, A., et al., Gradient-free federated learning methods with l 1 and l 2-randomization for non-smooth convex stochastic optimization problems, 2022.
10.
Zurück zum Zitat Gasnikov, A., et al., The power of first-order smooth optimization for black-box non-smooth problems, Proc. Int. Conf. Machine Learning, 2022, pp. 7241–7265. Gasnikov, A., et al., The power of first-order smooth optimization for black-box non-smooth problems, Proc. Int. Conf. Machine Learning, 2022, pp. 7241–7265.
11.
Zurück zum Zitat Bach, F. and Perchet, V., Highly-smooth zero-th order online optimization, Proc. Conf. Learning Theory, 2016, pp. 257–283. Bach, F. and Perchet, V., Highly-smooth zero-th order online optimization, Proc. Conf. Learning Theory, 2016, pp. 257–283.
12.
Zurück zum Zitat Beznosikov, A., Novitskii, V., and Gasnikov, A., One-point gradient-free methods for smooth and non-smooth saddle-point problems, Proc. 20th Int. Conf. Mathematical Optimization Theory and Operations Research (MOTOR), Irkutsk, Russia, 2021, pp. 144–158. Beznosikov, A., Novitskii, V., and Gasnikov, A., One-point gradient-free methods for smooth and non-smooth saddle-point problems, Proc. 20th Int. Conf. Mathematical Optimization Theory and Operations Research (MOTOR), Irkutsk, Russia, 2021, pp. 144–158.
13.
Zurück zum Zitat Akhavan, A., Pontil, M., and Tsybakov, A., Exploiting higher order smoothness in derivative-free optimization and continuous bandits, Adv. Neural Inf. Process. Syst., 2020, vol. 33, pp. 9017–9027. Akhavan, A., Pontil, M., and Tsybakov, A., Exploiting higher order smoothness in derivative-free optimization and continuous bandits, Adv. Neural Inf. Process. Syst., 2020, vol. 33, pp. 9017–9027.
14.
Zurück zum Zitat Polyak, B.T., Gradient methods for the minimisation of functionals, USSR Comput. Math. Math. Phys., 1963, vol. 3, no. 4, pp. 864–878.CrossRefMATH Polyak, B.T., Gradient methods for the minimisation of functionals, USSR Comput. Math. Math. Phys., 1963, vol. 3, no. 4, pp. 864–878.CrossRefMATH
15.
Zurück zum Zitat Łojasiewicz, S., Une propriété topologique des sous-ensembles analytiques réels, Les Equations aux Dérivées Partielles, 1963, vol. 117, pp. 87–89. Łojasiewicz, S., Une propriété topologique des sous-ensembles analytiques réels, Les Equations aux Dérivées Partielles, 1963, vol. 117, pp. 87–89.
16.
Zurück zum Zitat Ajalloeian, A. and Stich, S.U., On the convergence of SGD with biased gradients, 2020. Ajalloeian, A. and Stich, S.U., On the convergence of SGD with biased gradients, 2020.
17.
Zurück zum Zitat Lobanov, A., Gasnikov, A., and Stonyakin, F., Highly smoothness zero-order methods for solving optimization problems under PL condition, 2023. Lobanov, A., Gasnikov, A., and Stonyakin, F., Highly smoothness zero-order methods for solving optimization problems under PL condition, 2023.
18.
Zurück zum Zitat Yue, P., Fang, C., and Lin, Z., On the lower bound of minimizing Polyak–Łojasiewicz functions, 2022. Yue, P., Fang, C., and Lin, Z., On the lower bound of minimizing Polyak–Łojasiewicz functions, 2022.
19.
Zurück zum Zitat Yang, J., Kiyavash, N., and He, N., Global convergence and variance-reduced optimization for a class of nonconvex-nonconcave minimax problems, 2020. Yang, J., Kiyavash, N., and He, N., Global convergence and variance-reduced optimization for a class of nonconvex-nonconcave minimax problems, 2020.
20.
Zurück zum Zitat Akhavan, A., et al., Gradient-free optimization of highly smooth functions: Improved analysis and a new algorithm, 2023. Akhavan, A., et al., Gradient-free optimization of highly smooth functions: Improved analysis and a new algorithm, 2023.
21.
Zurück zum Zitat Nouiehed, M., et al., Solving a class of non-convex min-max games using iterative first order methods, Adv. Neural Inf. Process. Syst., 2019, vol. 32. Nouiehed, M., et al., Solving a class of non-convex min-max games using iterative first order methods, Adv. Neural Inf. Process. Syst., 2019, vol. 32.
23.
Zurück zum Zitat Beckner, W., A generalized Poincaré inequality for Gaussian measures, Proc. Am. Math. Soc., 1989, vol. 105, no. 2, pp. 397–400.MathSciNetMATH Beckner, W., A generalized Poincaré inequality for Gaussian measures, Proc. Am. Math. Soc., 1989, vol. 105, no. 2, pp. 397–400.MathSciNetMATH
24.
Zurück zum Zitat Karimi, H., Nutini, J., and Schmidt, M., Linear convergence of gradient and proximal-gradient methods under the Polyak–Łojasiewicz condition, Proc. Eur. Conf. Machine Learning and Knowledge Discovery in Databases (ECML PKDD), Riva del Garda, Italy, 2016, pp. 795–811. Karimi, H., Nutini, J., and Schmidt, M., Linear convergence of gradient and proximal-gradient methods under the Polyak–Łojasiewicz condition, Proc. Eur. Conf. Machine Learning and Knowledge Discovery in Databases (ECML PKDD), Riva del Garda, Italy, 2016, pp. 795–811.
Metadaten
Titel
Gradient-Free Algorithms for Solving Stochastic Saddle Optimization Problems with the Polyak–Łojasiewicz Condition
verfasst von
S. I. Sadykov
A. V. Lobanov
A. M. Raigorodskii
Publikationsdatum
01.12.2023
Verlag
Pleiades Publishing
Erschienen in
Programming and Computer Software / Ausgabe 6/2023
Print ISSN: 0361-7688
Elektronische ISSN: 1608-3261
DOI
https://doi.org/10.1134/S0361768823060063

Weitere Artikel der Ausgabe 6/2023

Programming and Computer Software 6/2023 Zur Ausgabe

Premium Partner