Noisy optimization with evolution strategies. With a foreword by Hans-Georg Beyer.
The book addresses one of the most pressing and interesting topics in evolutionary computation research -- the performance of evolutional algorithms in uncertain environments, i.e., applications where the objective functions are distributed by noise. Noise is a common phenomenon in many applications, even though it is mostly neglected or removed already at the stage of defining the objective function. When applying evolutionary algorithms, however, there is no need to remove the noise explicitly. Considering real-valued evolution strategies (ES), the author shows in the book under review that ES can deal with noise efficiently. That is, his theoretical results indicate that using a recombining population is more efficient than decreasing the noise by averaging objective function values (the usual procedure used in the optimization community). Furthermore, his investigations reveal a superiority of population-based recombinative general purpose ES over state-of the-direct optimization algorithms like simplex methods, response surface techniques, and implicit filtering which were explicitly designed for dealing with noise. After a brief Introduction chapter, the author introduces in Chapter 2 the \((\mu/\rho+\lambda)\)-ES with isotropic normal mutations and briefly discusses various mechanisms for the adaptation of the mutation strength. With the line and the sphere he then introduces two fitness environments that will be used throughout the book. Subsequently he turns to a brief discussion of local performance measures. Section 5 is of particular importance as it introduces a simple model for the spherical environment that forms the basis for most of the calculations in the following chapters. Chapter 3 summarize results from the analysis of the local performance of the \((1 +1)\)-ES in the spherical environment. The analysis differs from a previous study by Beyer in that it is not assumed that the fitness of surviving candidate solutions is reevaluated. The resulting overvaluation of the parental fitness is identified as a factor that strongly influences the performance of the strategy and that has largely been overlooked in the past. In Chapter 4, the local performance of the \((\mu,\lambda)\)-ES in the linear environment is studied. In contrast to the other strategies analyzed in this book, a population that is distributed in search space needs to be modeled, thereby greatly increasing the mathematical difficulties involved. At the same time, the results are of great interest as they can serve to explain the benefits of distributed populations of candidate solutions for optimization in the presence of noise. Chapter 5 starts with an analysis of the performance of the \((\mu/\mu,\lambda)\)-ES with intermediate recombination in the spherical environment. The results of analysis neatly generalize a number of previously known results and reveal how genetic resulting from recombination is especially beneficial in the presence of noise. However, while the results are strikingly simple, their accuracy is satisfactory only in very high-dimensional search spaces and small population sizes. Chapter 6 presents an empirical comparison of the performance of a number of search strategies on the noisy sphere. In addition to a \((\mu/\mu,\lambda)\)-ES with cumulative mutation strength adaptation, the direct pattern search algorithm of Hooke and Jeeves, the simplex algorithm of Nedler and Mead, the multi-directional search method of Torczon, the implicit filtering algorithm of Gilmore and Kelley, a variant of the implicit filtering algorithm that makes use of the simultaneous perturbation gradient approximation of Spall, and the evolutionary gradient search method of Salomon, are considered. Interesting differences in the scaling behaviors of the strategies are observed and strengths and weaknesses of the respective strategies are identified. Chapter 7 concludes with a summary and discussion of the main results and with goals for future research. Mathematical details have been omitted from many of the calculations and have been deferred to the appendix so as to increase the readability of the text and to minimize the technical difficulties for the reader mainly interested in results. Summing up, the book appears to be an interesting theoretical complement to many existing books describing practical applications of evolutionary computations.
- A comparison of evolution strategies with other direct search methods in the presence of noise
- Evolutionary algorithms in noisy environments: theoretical issues and guidelines for practise.
- Local performance of the (/_I,)-ES in a noisy environment
- scientific article; zbMATH DE number 1855686
- Evolution strategies in noisy environments. A survey of existing work
- Efficient covariance matrix update for variable metric evolution strategies
- Evolution strategies with cumulative step length adaptation on the noisy parabolic ridge
- Evolutionary algorithms in noisy environments: theoretical issues and guidelines for practise.
- Principles in noisy optimization. Applied to multi-agent coordination
- Performance analysis of evolution strategies with multi-recombination in high-dimensional R^N-search spaces disturbed by noise.
- A comparison of evolution strategies with other direct search methods in the presence of noise
- On the steady state analysis of covariance matrix self-adaptation evolution strategies on the noisy ellipsoid model
- Evolutionary optimisation of noisy multi-objective problems using confidence-based dynamic resampling
- Weighted multirecombination evolution strategies
- Analysing the robustness of evolutionary algorithms to noise: refined runtime bounds and an example where noise is beneficial
- Rare event probability estimation in the presence of epistemic uncertainty on input probability distribution parameters
- Evolution strategies in noisy environments. A survey of existing work
- Local performance of the (/_I,)-ES in a noisy environment
- Run-time analysis of population-based evolutionary algorithm in noisy environments
- An empirical study about the usefulness of evolution strategies to solve constrained optimization problems
- A Markov chain that models genetic algorithms in noisy environments
- Markov chain analysis of genetic algorithms applied to fitness functions perturbed concurrently by additive and multiplicative noise
- The steady state behavior of (/_I, )-ES on ellipsoidal fitness models disturbed by noise
- Selection in the presence of noise
- The theory of evolution strategies
- History and immortality in evolutionary computation
- scientific article; zbMATH DE number 1855686 (Why is no real title available?)
- Controlling population size and mutation strength by Meta-ES under fitness noise
- Noisy optimization complexity under locality assumption
- A multi-level optimization approach for the planning of heterogeneous sensor networks
- Noise analysis compact differential evolution
- Foundations of Genetic Algorithms
- Resampling vs recombination: a statistical run time estimation
- KI 2005: Advances in Artificial Intelligence
- Progress analysis of a multi-recombinative evolution strategy on the highly multimodal Rastrigin function
- Convergence Properties of the (μ/μI, λ)-ES on the Rastrigin Function
- Log-linear convergence and divergence of the scale-invariant \((1+1)\)-ES in noisy environments
- Effective matrix adaptation strategy for noisy derivative-free optimization
This page was built for publication: Noisy optimization with evolution strategies. With a foreword by Hans-Georg Beyer.
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4449264)
