Abstract
In this paper, we propose an asynchronous distributed learning algorithm where parameter updates are performed by worker machines simultaneously on a local sub-part of the training data. These workers send their updates to a master machine that coordinates all received parameters in order to minimize a global empirical loss. The communication exchanges between workers and the master machine are generally the bottleneck of most asynchronous scenarios. We propose to reduce this communication cost by a sparsification mechanism which, for each worker machine, consists in randomly and independently choosing some local update entries that will not be transmitted to the master. We provably show that if the probability of choosing such local entries is high and that the global loss is strongly convex, then the whole process is guaranteed to converge to the minimum of the loss. In the case where this probability is low, we empirically show on three datasets that our approach converges to the minimum of the loss in most of the cases with a better convergence rate and much less parameter exchanges between the master and the worker machines than without using our sparsification technique.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Bach, F., Jenatton, R., Mairal, J., Obozinski, G., et al.: Optimization with sparsity-inducing penalties. Found. Trends® Mach. Learn. 4(1), 1–106 (2012)
Boyd, S.P., Parikh, N., Chu, E., Peleato, B., Eckstein, J.: Distributed optimization and statistical learning via the alternating direction method of multipliers. Found. Trends Mach. Learn. 3(1), 1–122 (2011)
Candes, E.J., Wakin, M.B., Boyd, S.P.: Enhancing sparsity by reweighted l 1 minimization. J. Fourier Anal. Appl. 14(5–6), 877–905 (2008)
Chen, J., Monga, R., Bengio, S., Jozefowicz, R.: Revisiting distributed synchronous SGD. In: International Conference on Learning Representations Workshop Track (2016). https://arxiv.org/abs/1604.00981
Fadili, J., Malick, J., Peyré, G.: Sensitivity analysis for mirror-stratifiable convex functions. SIAM J. Optim. 28(4), 2975–3000 (2018)
Grishchenko, D., Iutzeler, F., Malick, J., Amini, M.R.: Asynchronous distributed learning with sparse communications and identification. arXiv preprint arXiv:1812.03871 (2018)
Hannah, R., Yin, W.: On unbounded delays in asynchronous parallel fixed-point algorithms. J. Sci. Comput. 76(1), 299–326 (2017). https://doi.org/10.1007/s10915-017-0628-z
Konečnỳ, J., McMahan, H.B., Ramage, D., Richtárik, P.: Federated optimization: distributed machine learning for on-device intelligence. arXiv:1610.02527 (2016)
Kumar, V.: Introduction to Parallel Computing. Addison-Wesley Longman (2002)
Lee, S., Wright, S.J.: Manifold identification in dual averaging for regularized stochastic online learning. J. Mach. Learn. Res. 13(1), 1705–1744 (2012)
Lin, Y., Han, S., Mao, H., Wang, Y., Dally, W.J.: Deep gradient compression: Reducing the communication bandwidth for distributed training. arXiv preprint arXiv:1712.01887 (2017)
Ma, C., Jaggi, M., Curtis, F.E., Srebro, N., Takáč, M.: An accelerated communication-efficient primal-dual optimization framework for structured machine learning. Optimization Methods and Software, pp. 1–25 (2019)
Mishchenko, K., Iutzeler, F., Malick, J.: A distributed flexible delay-tolerant proximal gradient algorithm. SIAM J. Optim. 30(1), 933–959 (2020)
Mishchenko, K., Iutzeler, F., Malick, J., Amini, M.R.: A delay-tolerant proximal-gradient algorithm for distributed learning. In: Proceedings of the 35th International Conference on Machine Learning (ICML), vol. 80, pp. 3587–3595 (2018)
Nutini, J., Schmidt, M., Hare, W.: “active-set complexity” of proximal gradient: how long does it take to find the sparsity pattern? Optimization Lett. 13(4), 645–655 (2019)
Sun, T., Hannah, R., Yin, W.: Asynchronous coordinate descent under more realistic assumptions. In: Advances in Neural Information Processing Systems (2017)
Valiant, L.G.: A bridging model for parallel computation. Commun. ACM 33(8), 103–111 (1990)
Wangni, J., Wang, J., Liu, J., Zhang, T.: Gradient sparsification for communication-efficient distributed optimization. In: Advances in Neural Information Processing Systems, pp. 1306–1316 (2018)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
1 Electronic supplementary material
Below is the link to the electronic supplementary material.
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this paper
Cite this paper
Grischenko, D., Iutzeler, F., Amini, MR. (2020). Sparse Asynchronous Distributed Learning. In: Yang, H., Pasupa, K., Leung, A.CS., Kwok, J.T., Chan, J.H., King, I. (eds) Neural Information Processing. ICONIP 2020. Communications in Computer and Information Science, vol 1333. Springer, Cham. https://doi.org/10.1007/978-3-030-63823-8_50
Download citation
DOI: https://doi.org/10.1007/978-3-030-63823-8_50
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-63822-1
Online ISBN: 978-3-030-63823-8
eBook Packages: Computer ScienceComputer Science (R0)