Abstract
We derive a stochastic search procedure for parameter optimization from two first principles: (1) imposing the least prior assumptions, namely by maximum entropy sampling, unbiasedness and invariance; (2) exploiting all available information under the constraints imposed by (1). We additionally require that two of the most basic functions can be solved reasonably fast. Given these principles, two principal heuristics are used: reinforcing of good solutions and good steps (increasing their likelihood) and rendering successive steps orthogonal. The resulting search algorithm is the covariance matrix adaptation evolution strategy, CMA-ES, that coincides to a great extent to a natural gradient descent. The invariance properties of the CMA-ES are formalized, as are its maximum likelihood and stationarity properties. A small parameter study for a specific heuristic—deduced from the principles of reinforcing good steps and exploiting all information—is presented, namely for the cumulation of an evolution or search path. Experiments on two noisy functions are provided.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
That is, to find a sequence θ k , \(k = 1, 2, 3,\ldots\), such that \(\lim _{k\rightarrow \infty }E(f(\boldsymbol{x}\vert \theta _{k})) = {f}^{{\ast}}\).
- 2.
Different learning rates might be related to some parameters in the distribution being orthogonal.
- 3.
In the (μ, λ)-ES, only the μ best samples are selected for the next iteration. Given μ = 1, a very general optimality condition for λ states that the currently second best solution must resemble the f-value of the previous best solution [24]. Consequently, on any linear function, λ = 2 and λ = 3 are optimal [24, 36]. On the sphere function Eq. (8.22), λ = 5 is optimal [33]. On the latter, \(\lambda \approx 3.7\mu\) can also be shown optimal for μ ≥ 2 and equal recombination weights [9], compare (8.12). For λ < 5, the original strategy parameter setting for CMA-ES has been rectified in [10], but only mirrored sampling leads to satisfactory performance in this case [10].
- 4.
The positive symmetric square root satisfies \({\boldsymbol{C}_{k}}^{-\frac{1} {2} }{\boldsymbol{C}_{ k}}^{-\frac{1} {2} } ={ \boldsymbol{C}_{ k}}^{-1}\), has only positive eigenvalues and is unique.
- 5.
Source code is available at http://www.lri.fr/~hansen/cmaes_inmatlab.html and will be accessible at http://cma.gforge.inria.fr/ in the future. In our experiment, version 3.40.beta was used with Matlab.
- 6.
There is a simple way to smooth the landscape: A single evaluation can be replaced by the median (not the mean) of a number of evaluations. Only a few evaluations reduce the dispersion considerably, but about 1, 000 evaluations are necessary to render the landscape similarly smooth as with α N = 0. 01. Together with (\(\mu /\mu _{\mathrm{}\,{w}}\))-CMA-ES, single evaluations, as in Fig. 8.10, need overall the least number of function evaluations (comprising restarts).
References
Y. Akimoto, Y. Nagata, I. Ono, S. Kobayashi, Bidirectional relation between CMA evolution strategies and natural evolution strategies, in Proceedings of the Parallel Problem Solving from Nature – PPSN XI, Part I, Kraków, ed. by R. Schaefer, C. Cotta, J. Kolodziej, G. Rudolph. Lecture Notes in Computer Science, vol. 6238 (Springer, 2010), pp. 154–163
D. Arnold, Optimal weighted recombination, in Foundations on Genetic Algorithms FOGA 2005, Aizu-Wakamatsu City. Lecture Notes in Computer Science, vol. 3469 (Springer, 2005), pp. 215–237
D. Arnold, Weighted multirecombination evolution strategies. Theor. Comput. Sci. 361(1), 18–37 (2006)
D.V. Arnold, N. Hansen, Active covariance matrix adaptation for the (1+1)-CMA-ES, in Proceedings of the Genetic and Evolutionary Computation Conference, GECCO 2010, Portland, 2010, pp. 385–392
L. Arnold, A. Auger, N. Hansen, Y. Ollivier, Information-geometric optimization algorithms: a unifying picture via invariance principles. arXiv:1106.3708 (Arxiv preprint) (2011)
A. Auger, N. Hansen, A restart CMA evolution strategy with increasing population size, in The 2005 IEEE International Congress on Evolutionary Computation (CEC 2005), Edinburgh, ed. by B. McKay et al., vol. 2, 2005, pp. 1769–1776
A.Auger, N. Hansen, Reconsidering the progress rate theory for evolution strategies in finite dimensions, in Proceedings of the 8th Annual Conference on Genetic and Evolutionary Computation GECCO, Seattle (ACM, 2006), pp. 445–452
A. Auger, N. Hansen, J. Zerpa, R. Ros, M. Schoenauer, Experimental comparisons of derivative free optimization algorithms, in 8th International Symposion on Experimental Algorithms SEA 2009, Dortmund. Lecture Notes in Computer Science, vol. 5526 (Springer, 2009), pp. 3–15
H.G. Beyer, The Theory of Evolution Strategies. Natural Computing Series (Springer, Heidelberg, 2001)
D. Brockhoff, A. Auger, N. Hansen, D.V. Arnold, T. Hohm, Mirrored sampling and sequential selection for evolution strategies, in Parallel Problem Solving from Nature (PPSN XI), Kraków, ed. by R. Schaefer et al. LNCS, vol. 6238 (Springer, 2010) pp. 11–20
T. Glasmachers, T. Schaul, Y. Sun, D. Wierstra, J. Schmidhuber, Exponential natural evolution strategies, in Proceedings of the Genetic and Evolutionary Computation Conference, GECCO 2010, Portland, ed. by M. Pelikan, J. Branke (ACM, 2010) pp. 393–400
N. Hansen, The CMA evolution strategy: a tutorial, http://www.lri.fr/~hansen/cmatutorial.pdf
N. Hansen, Invariance, self-adaptation and correlated mutations in evolution strategies, in Proceedings of PPSN VI, Parallel Problem Solving from Nature, Paris, ed. by M. Schoenauer, K. Deb, G. Rudolph, X. Yao, E. Lutton, J. Merelo, H.P. Schwefel (Springer, 2000), pp. 355–364
N. Hansen, An analysis of mutative σ-self-adaptation on linear fitness functions. Evol. Comput. 14(3), 255–275 (2006)
N. Hansen, The CMA evolution strategy: a comparing review, in Towards a New Evolutionary Computation. Advances on Estimation of Distribution Algorithms, Hefei, ed. by J. Lozano, P. Larranaga, I. Inza, E. Bengoetxea (Springer, 2006), pp. 75–102
N. Hansen, Adaptive encoding for optimization. Research Report RR-6518, INRIA (2008), http://hal.inria.fr/inria-00275983/en/
N. Hansen, Adaptive encoding: how to render search coordinate system invariant, in Parallel Problem Solving from Nature (PPSN X), Dortmund, ed. by G. Rudolph et al. LNCS, 2008, pp. 205–214
N. Hansen, CMA-ES with two-point step-size adaptation. Technical Report, RR-6527, INRIA (2008), http://hal.inria.fr/inria-00276854/en/
N. Hansen, Benchmarking a BI-population CMA-ES on the BBOB-2009 function testbed, in Workshop Proceedings of the GECCO Genetic and Evolutionary Computation Conference, Montreal (ACM, 2009), pp. 2389–2395
N. Hansen, Benchmarking a BI-population CMA-ES on the BBOB-2009 noisy testbed, in Workshop Proceedings of the GECCO Genetic and Evolutionary Computation Conference, Montreal (ACM, 2009), pp. 2397–2402
N. Hansen, S. Kern, Evaluating the CMA evolution strategy on multimodal test functions, in Parallel Problem Solving from Nature PPSN VIII, Birmingham, ed. by X. Yao, et al. Lecture Notes in Computer Science, vol. 3242 (Springer, 2004), pp. 282–291
N. Hansen, A. Ostermeier, Completely derandomized self-adaptation in evolution strategies. Evol. Comput. 9(2), 159–195 (2001)
N. Hansen, R. Ros, Benchmarking a weighted negative covariance matrix update on the BBOB-2010 noiseless testbed, in Genetic and Evolutionary Computation Conference, GECCO 2010, Companion Material, Portland, 2010, pp. 1673–1680
N. Hansen, A. Gawelczyk, A. Ostermeier, Sizing the population with respect to the local progress in (1, λ)-evolution strategies—a theoretical analysis, in IEEE International Conference on Evolutionary Computation, Perth, vol. 1, 1995, pp. 80–85
N. Hansen, S. Niederberger, L. Guzzella, P. Koumoutsakos, A method for handling uncertainty in evolutionary optimization with an application to feedback control of combustion. IEEE Trans. Evol. Comput. 13(1), 180–197 (2009)
N. Hansen, A. Auger, R. Ros, S. Finck, P. Pošík, Comparing results of 31 algorithms from the black-box optimization benchmarking BBOB-2009, in Workshop Proceedings of the Genetic and Evolutionary Computation Conference (GECCO 2010), Portland (ACM, 2010), pp. 1689–1696
J. Jägersküpper, Lower bounds for hit-and-run direct search, in Stochastic Algorithms: Foundations and Applications – SAGA 2007, Zurich, ed. by Yao, Xin et al. LNCS, vol. 4665 (Springer, Berlin/Heidelberg, 2007) pp. 118–129
J. Jägersküpper, Lower bounds for randomized direct search with isotropic sampling. Oper. Res. Lett. 36(3), 327–332 (2008)
G. Jastrebski, D. Arnold, Improving evolution strategies through active covariance matrix adaptation, in The 2006 IEEE International Congress on Evolutionary Computation (CEC 2006), Vancouver, 2006, pp. 2814–2821
M. Jebalia, A. Auger, N. Hansen, Log linear convergence and divergence of the scale-invariant (1+1)-ES in noisy environments. Algorithmica 59(3), 425–460 (2011)
T. Jones, S. Forrest, Fitness distance correlation as a measure of problem difficulty for genetic algorithms, in Proceedings of the 6th International Conference on Genetic Algorithms, ICGA, Pittsburgh, ed. by L.J. Eshelman (Morgan Kaufmann, 1995), pp. 184–192
A. Ostermeier, A. Gawelczyk, N. Hansen, Step-size adaptation based on non-local use of selection information, in Parallel Problem Solving from Nature PPSN IV, Jerusalem, ed. by Y. Davidor et al. Lecture Notes in Computer Science, vol. 866 (Springer, 1994), pp. 189–198
I. Rechenberg, Evolutionsstrategie: Optimierung technischer Systeme nach Prinzipien der biologischen Evolution (Frommann-Holzboog, Stuttgart, 1973)
R. Salomon, J.L. van Hemmen, Accelerating backpropagation through dynamic self-adaptation. Neural Netw. 9(4), 589–601 (1996)
M. Schumer, K. Steiglitz, Adaptive step size random search. IEEE Trans. Autom. Control 13(3), 270–276 (1968)
H.P. Schwefel, Numerical Optimization of Computer Models (Wiley, New York, 1981)
T. Suttorp, N. Hansen, C. Igel, Efficient covariance matrix update for variable metric evolution strategies. Mach. Learn. 75(2), 167–197 (2009)
O. Teytaud, H. Fournier, Lower bounds for evolution strategies using VC-dimension, in Parallel Problem Solving from Nature PPSN X, Dortmund. Lecture Notes in Computer Science, vol. 5199 (Springer, 2008) pp. 102–111
O. Teytaud, S. Gelly, General lower bounds for evolutionary algorithms, in Parallel Problem Solving from Nature PPSN IX, Reykjavik. Lecture Notes in Computer Science, vol. 4193 (Springer, 2006), pp. 21–31
D. Wierstra, T. Schaul, J. Peters, J. Schmidhuber, Natural evolution strategies, in IEEE Congress on Evolutionary Computation, Hong Kong (IEEE, 2008), pp. 3381–3387
Acknowledgements
The authors would like to express their gratitude to Marc Schoenauer for his kind and consistent support.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2014 Springer-Verlag Berlin Heidelberg
About this chapter
Cite this chapter
Hansen, N., Auger, A. (2014). Principled Design of Continuous Stochastic Search: From Theory to Practice. In: Borenstein, Y., Moraglio, A. (eds) Theory and Principled Methods for the Design of Metaheuristics. Natural Computing Series. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-33206-7_8
Download citation
DOI: https://doi.org/10.1007/978-3-642-33206-7_8
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-642-33205-0
Online ISBN: 978-3-642-33206-7
eBook Packages: Computer ScienceComputer Science (R0)