Abstract
The maximum function, on vectors of real numbers, is not differentiable. Consequently, several differentiable approximations of this function are popular substitutes. We survey three smooth functions which approximate the maximum function and analyze their convergence rates. We interpret these functions through the lens of tropical geometry, where their performance differences are geometrically salient. As an application, we provide an algorithm which computes the max-convolution of two integer vectors in quasi-linear time. We show this algorithm’s power in computing adjacent sums within a vector as well as computing service curves in a network analysis application.
Similar content being viewed by others
Data availability
All data is described in the text.
References
Zang, I.: A smoothing-out technique for min-max optimization. Math. Programming. 19(1), 61–77 (1980)
Zhao, G., Wang, Z., Mou, H.: Uniform approximation of min/max functions by smooth splines. J. Comput. Appl. Math. 236(5), 699–703 (2011). The 7th International Conference on Scientific Computing and Applications, June 13–16, 2010, Dalian, China
Asadi, K., Littman, M.L.: An alternative softmax operator for reinforcement learning. In: Proceedings of the 34th International Conference on Machine Learning, Vol 70. ICML’17, pp. 243–252. JMLR.org, (Online) (2017)
Liu, W., Wen, Y., Yu, Z., Yang, M.: Large-margin softmax loss for convolutional neural networks. In: Balcan, M., Weinberger, K. (eds.) International Conference on Machine Learning, Vol 48. Proceedings of Machine Learning Research, vol. 48 (2016). 33rd International Conference on Machine Learning, New York, NY, JUN 20–22, 2016
Nielsen, F., Sun, K.: Guaranteed bounds on information-theoretic measures of univariate mixtures using piecewise log-sum-exp inequalities. Entropy. 18(12) (2016). https://doi.org/10.3390/e18120442
Blanchard, P., Higham, D.J., Higham, N.J.: Accurately computing the log-sum-exp and softmax functions. IMA J. Numer. Anal. 41(4), 2311–2330 (2020). https://doi.org/10.1093/imanum/draa038. https://academic.oup.com/imajna/articlepdf/41/4/2311/40758053/draa038.pdf
Trefethen, L.N., Weideman, J.A.C.: The exponentially convergent trapezoidal rule. SIAM Review. 56(3), 385–458 (2014)
Pfeuffer, J., Serang, O.: A bounded p-norm approximation of max-convolution for sub-quadratic Bayesian inference on additive factors. J. Mach. Learn. Res. 17, 36–13639 (2016)
Serang, O.: A fast numerical method for max-convolution and the application to efficient max-product inference in bayesian networks. J. Comput. Biol. J. Mol. Cell. Biol. 22 (2015). https://doi.org/10.1089/cmb.2015.0013
Maclagan, D., Sturmfels, B.: Introduction to tropical geometry. Graduate Studies in Mathematics, vol. 161. American Mathematical Society, Providence, Rhode Island (2015)
Bergman, G.M.: The logarithmic limit-set of an algebraic variety. Trans. Am. Math. Soc. 157, 459–469 (1971)
DeWolff, T., Schroeter, F.: The boundary of amoebas. arXiv Preprint: - arXiv:1310.7363 (2013)
Forsberg, M., Passare, M., Tsikh, A.: Laurent determinants and arrangements of hyperplane amoebas. Adv. Math. 151, 45–70 (2000)
Moenck, R.T.: Practical fast polynomial multiplication. In: Proceedings of the Third ACM Symposium on Symbolic and Algebraic Computation. SYMSAC’ 76, pp. 136–148. Association for Computing Machinery, New York, NY, USA (1976)
Cygan, M., Mucha, M., Wundefinedgrzycki, K., Włodarczyk, M.: On problems equivalent to (min,+)-convolution. ACM Trans. Algorithms 15(1) (2019)
Van Bemten, A., Kellerer, W.: Network calculus: a comprehensive guide (2016). https://doi.org/10.13140/RG.2.2.32305.89448
Funding
This material is based upon work supported by the National Science Foundation under grants CCF-1812746 and CMMI-2041789. The first author is partially supported by an NSERC Discovery Grant (Canada).
Author information
Authors and Affiliations
Contributions
J. D. H. and T. B. developed the theories and algorithms and wrote the main manuscript. T. B. and C. H. tested the convergence rates; T.B. prepared Figs. 1–13 and 16–18. T.B. analyzed and wrote Section 4 regarding tropical geometry and wrote Section 6.1. C.H. tested and prepared Figs. 14–15 and 19–20 and wrote Sects. 5.1–5.2 and 6.2. All authors reviewed the manuscript.
Corresponding author
Ethics declarations
Ethical approval
Not applicable
Conflict of interest
The authors declare no competing interests.
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
Brysiewicz, T., Hauenstein, J.D. & Hills, C. Max-convolution through numerics and tropical geometry. Numer Algor (2023). https://doi.org/10.1007/s11075-023-01668-w
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s11075-023-01668-w