Abstract
Stochastic composite objective mirror descent (SCOMID) is an effective method for solving large-scale stochastic composite problems in machine learning. This method can efficiently use the geometric properties of a problem through a general distance function. However, most existing analyses rely on the convexity of the problem and the unbiased assumption of the stochastic gradient. In addition, the convergence results are obtained in expectation. To this end, we present an almost sure convergence analysis of SCOMID with biased gradient estimation in the non-convex non-smooth setting. For this general case, the analysis shows that the minimum of the squared generalized projected gradient norm arbitrarily converges to zero with probability one. We also obtain the almost sure convergence of function values for SCOMID with time-varying stepsizes in the non-convex and non-smooth setting. Numerical experiments support our theoretical findings.
Similar content being viewed by others
Data Availibility Statement
Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.
Notes
\({\mathcal {F}}_{t}\) denotes the \(\sigma\)-algebra generated by random variables \(x_{1}\), \(x_{2}\), \(\cdots\), \(x_{t}\), i.e., \({\mathcal {F}}_{t}=\sigma (x_{1},\,x_{2},\,\cdots ,\,x_{t})\).
\(f(x)\sim g(x)\): there exist \(x_{0}\), such that \(\lim _{x\rightarrow x_{0}} f(x)/g(x)=1\).
LIBSVM website: https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/.
References
Alacaoglu, A., Malitsky, Y., Cevher, V.: Convergence of adaptive algorithms for weakly convex constrained optimization. arXiv preprint arXiv:2006.06650 (2020)
Atchadé, Y.F., Fort, G., Moulines, E.: On perturbed proximal gradient algorithms. J. Mach. Learn. Res. 18(1), 310–342 (2017)
Bottou, L., Curtis, F.E., Nocedal, J.: Optimization methods for large-scale machine learning. SIAM Rev. 60(2), 223–311 (2018)
Cevher, V., Vũ, B.C.: On the linear convergence of the stochastic gradient method with constant step-size. Optim. Lett. 13(5), 1177–1187 (2019)
Davis, D., Drusvyatskiy, D.: Stochastic model-based minimization of weakly convex functions. SIAM J. Optim. 29(1), 207–239 (2019)
Driggs, D., Liang, J., Schönlieb, C.B.: On biased stochastic gradient estimation. J. Mach. Learn. Res. 23, 24–1 (2022)
Drusvyatskiy, D., Paquette, C.: Efficiency of minimizing compositions of convex functions and smooth maps. Math. Program. 178(1), 503–558 (2019)
Duchi, J., Hazan, E., Singer, Y.: Adaptive subgradient methods for online learning and stochastic optimization. J. Mach. Learn. Res. 12(7), 2121–2159 (2011)
Duchi, J., Shalev-Shwartz, S., Singer, Y., Tewari, A.: Composite objective mirror descent. In: Conference on Learning Theory, vol. 10, pp. 14–26 (2010)
Duchi, J., Singer, Y.: Efficient online and batch learning using forward backward splitting. J. Mach. Learn. Res. 10, 2899–2934 (2009)
Dundar, M., Krishnapuram, B., Bi, J., Rao, R.B.: Learning classifiers when the training data is not iid. In: IJCAI, vol. 2007, pp. 756–61 (2007)
Ghadimi, S., Lan, G., Zhang, H.: Mini-batch stochastic approximation methods for nonconvex stochastic composite optimization. Math. Program. 155(1), 267–305 (2016)
Gorbunov, E., Hanzely, F., Richtárik, P.: A unified theory of SGD: Variance reduction, sampling, quantization and coordinate descent. In: International Conference on Artificial Intelligence and Statistics, pp. 680–690. PMLR (2020)
Gower, R., Sebbouh, O., Loizou, N.: SGD for structured nonconvex functions: Learning rates, minibatching and interpolation. In: International Conference on Artificial Intelligence and Statistics, pp. 1315–1323. PMLR (2021)
Han, Y., Feng, X., Baciu, G., Wang, W.: Nonconvex sparse regularizer based speckle noise removal. Pattern Recognit. 46(3), 989–1001 (2013)
J. Reddi, S., Sra, S., Poczos, B., Smola, A.J.: Proximal stochastic methods for nonsmooth nonconvex finite-sum optimization. In: Advances in Neural Information Processing Systems, vol. 29, pp. 1153–1161 (2016)
Khaled, A., Richtárik, P.: Better theory for SGD in the nonconvex world. arXiv preprint arXiv:2002.03329 (2020)
Li, Z., Li, J.: Simple and optimal stochastic gradient methods for nonsmooth nonconvex optimization. J. Mach. Learn. Res. 23(239), 1–61 (2022)
Liang, J., Monteiro, R.D.: An average curvature accelerated composite gradient method for nonconvex smooth composite optimization problems. SIAM J. Optim. 31(1), 217–243 (2021)
Liu, J., Kong, J., Xu, D., Qi, M., Lu, Y.: Convergence analysis of AdaBound with relaxed bound functions for non-convex optimization. Neural Netw. 145, 300–307 (2022)
Liu, J., Yuan, Y.: On almost sure convergence rates of stochastic gradient methods. arXiv preprint arXiv:2202.04295 (2022)
Luo, J., Liu, J., Xu, D., Zhang, H.: SGD-r\(\alpha\): A real-time \(\alpha\)-suffix averaging method for SGD with biased gradient estimates. Neurocomputing 487, 1–8 (2022)
Mai, V., Johansson, M.: Convergence of a stochastic gradient method with momentum for non-smooth non-convex optimization. In: International Conference on Machine Learning, pp. 6630–6639. PMLR (2020)
Mandic, D., Chambers, J.: Recurrent Neural Networks for Prediction: Learning Algorithms, Architectures and Stability. Wiley, New York (2001)
Mertikopoulos, P., Hallak, N., Kavis, A., Cevher, V.: On the almost sure convergence of stochastic gradient descent in non-convex problems. In: Advances in Neural Information Processing Systems, vol. 33, pp. 1117–1128 (2020)
Metel, M.R., Takeda, A.: Stochastic proximal methods for non-smooth non-convex constrained sparse optimization. J. Mach. Learn. Res. 22, 115–1 (2021)
Nesterov, Y.: Lectures on Convex Optimization, vol. 137. Springer, Cham (2018)
Nikolova, M., Ng, M.K., Tam, C.P.: Fast nonconvex nonsmooth minimization methods for image restoration and reconstruction. IEEE Trans. Image Process. 19(12), 3073–3088 (2010)
Robbins, H., Monro, S.: A stochastic approximation method. Ann. Math. Stat. pp. 400–407 (1951)
Robbins, H., Siegmund, D.: A convergence theorem for non-negative almost supermartingales and some applications. In: Optimizing methods in statistics, pp. 233–257. Elsevier (1971)
Shalev-Shwartz, S., Ben-David, S.: Understanding Machine Learning: from Theory to Algorithms. Cambridge University Press, New York (2014)
Sun, R.Y.: Optimization for deep learning: An overview. J. Oper. Res. Soc. China 8(2), 249–294 (2020)
Tao, W., Pan, Z., Wu, G., Tao, Q.: Primal averaging: A new gradient evaluation step to attain the optimal individual convergence. IEEE T. Cybern. 50(2), 835–845 (2018)
Vial, J.P.: Strong and weak convexity of sets and functions. Math. Oper. Res. 8(2), 231–259 (1983)
Ward, R., Wu, X., Bottou, L.: AdaGrad stepsizes: Sharp convergence over nonconvex landscapes. J. Mach. Learn. Res. 21, 1–30 (2020)
Wood, K., Bianchin, G., Dall’Anese, E.: Online projected gradient descent for stochastic optimization with decision-dependent distributions. IEEE Control Syst. Lett. 6, 1646–1651 (2022)
Zhang, H., Pan, L., Xiu, N.: Optimality conditions for locally Lipschitz optimization with \(l_{0}\)-regularization. Optim. Lett. 15(1), 189–203 (2021)
Zhou, D., Chen, J., Cao, Y., Tang, Y., Yang, Z., Gu, Q.: On the convergence of adaptive gradient methods for nonconvex optimization. arXiv preprint arXiv:1808.05671 (2018)
Zhou, Y., Wang, Z., Ji, K., Liang, Y., Tarokh, V.: Proximal gradient algorithm with momentum and flexible parameter restart for nonconvex optimization. arXiv preprint arXiv:2002.11582 (2020)
Acknowledgements
The authors wish to thank the anonymous reviewers for their insightful and very helpful expert comments and suggestions. This work was funded in part by National Key R &D Program of China (No. 2021YFA1003400), in part by the National Natural Science Foundation of China (No. 62176051), and in part by the Fundamental Research Funds for the Central Universities of China (No. 2412020FZ024).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Liang, Y., Xu, D., Zhang, N. et al. Almost sure convergence of stochastic composite objective mirror descent for non-convex non-smooth optimization. Optim Lett (2023). https://doi.org/10.1007/s11590-023-01972-3
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s11590-023-01972-3