Abstract
It is known that step size adaptive evolution strategies (ES) do not converge (prematurely) to regular points of continuously differentiable objective functions. Among critical points, convergence to minima is desired, and convergence to maxima is easy to exclude. However, surprisingly little is known on whether ES can get stuck at a saddle point. In this work we establish that even the simple (1+1)-ES reliably overcomes most saddle points under quite mild regularity conditions. Our analysis is based on drift with tail bounds. It is non-standard in that we do not even aim to estimate hitting times based on drift. Rather, in our case it suffices to show that the relevant time is finite with full probability.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
Notes
- 1.
It should be noted that a few interesting cases exist for zero eigenvalues (which should be improbable in practice), like the “Monkey saddle” \(f(x) = x_1^3 - 3 x_1 x_2^2\). We believe that this case can be analyzed with the same techniques as developed below, but it is outside the scope of this paper.
References
Akimoto, Y.: Saddle point optimization with approximate minimization oracle. Tech. Rep. 2103.15985 (2021). arXiv.org
Akimoto, Y., Auger, A., Glasmachers, T.: Drift theory in continuous search spaces: expected hitting time of the (1+1)-es with 1/5 success rule. In: Proceedings of the Genetic and Evolutionary Computation Conference (GECCO). ACM (2018)
Akimoto, Y., Nagata, Y., Ono, I., Kobayashi, S.: Theoretical analysis of evolutionary computation on continuously differentiable functions. In: Genetic and Evolutionary Computation Conference, pp. 1401–1408. ACM (2010)
Dauphin, Y.N., Pascanu, R., Gulcehre, C., Cho, K., Ganguli, S., Bengio, Y.: Identifying and attacking the saddle point problem in high-dimensional non-convex optimization. In: Ghahramani, Z., Welling, M., Cortes, C., Lawrence, N.D., Weinberger, K.Q. (eds.) Advances in Neural Information Processing Systems, vol. 27, pp. 2933–2941 (2014)
Glasmachers, T.: Global convergence of the (1+1) evolution strategy. Evol. Comput. J. (ECJ) 28(1), 27–53 (2020)
Hajek, B.: Hitting-time and occupation-time bounds implied by drift analysis with applications. Adv. Appli. Prob. 14(3), 502–525 (1982)
Hansen, N., Ostermeier, A.: Completely derandomized self-adaptation in evolution strategies. Evol. Comput. 9(2), 159–195 (2001)
Kern, S., Müller, S.D., Hansen, N., Büche, D., Ocenasek, J., Koumoutsakos, P.: Learning probability distributions in continuous evolutionary algorithms-a comparative review. Nat. Comput. 3(1), 77–112 (2004)
Lehre, P.K., Witt, C.: General drift analysis with tail bounds. Tech. Rep. 1307.2559 (2013). arXiv.org
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2022 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Glasmachers, T. (2022). The (1+1)-ES Reliably Overcomes Saddle Points. In: Rudolph, G., Kononova, A.V., Aguirre, H., Kerschke, P., Ochoa, G., Tušar, T. (eds) Parallel Problem Solving from Nature – PPSN XVII. PPSN 2022. Lecture Notes in Computer Science, vol 13399. Springer, Cham. https://doi.org/10.1007/978-3-031-14721-0_22
Download citation
DOI: https://doi.org/10.1007/978-3-031-14721-0_22
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-14720-3
Online ISBN: 978-3-031-14721-0
eBook Packages: Computer ScienceComputer Science (R0)