Sparse Covariance Matrix Adaptation Techniques for Evolution Strategies

Meyer-Nieberg, Silja; Kropat, Erik

doi:10.1007/978-3-319-25032-8_1

Cited by 4 publications

(16 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The current analysis extends the work carried out in (Meyer-Nieberg and Kropat, 2014;Meyer-Nieberg and Kropat, 2015a) and augments the investigation conducted in (Meyer-Nieberg and Kropat, 2015b;Meyer-Nieberg and Kropat, 2015c) for the case of thresholding estimators. (Meyer-Nieberg and Kropat, 2014;Meyer-Nieberg and Kropat, 2015a) presented the first approaches to apply Ledoit-Wolf shrinkage estimators in evolution strategies.…”

Section: Introductionsupporting

confidence: 72%

Evolution Strategies and Covariance Matrix Adaptation - Investigating New Shrinkage Techniques

Meyer-Nieberg

Kropat

2016

Proceedings of the 8th International Conference on Agents and Artificial Intelligence

Self Cite

View full text Add to dashboard Cite

This paper discusses the covariance matrix adaptation in evolution strategies, a central and essential mechanism for the search process. Basing the estimation of the covariance matrix on small samples w.r.t. the search space dimension is known to be problematic. However, this situation is common in optimization raising the question, whether the performance of the evolutionary algorithms could be improved. In statistics, several approaches have been developed recently to improve the quality of the maximum-likelihood estimate. However, they are seldom applied in evolutionary computation. Here, we focus on linear shrinkage which requires relatively little additional effort. Several approaches and shrinkage targets are integrated into evolution strategies and analyzed in a series of experiments.

show abstract

Section: Introductionsupporting

confidence: 72%

Evolution Strategies and Covariance Matrix Adaptation - Investigating New Shrinkage Techniques

Meyer-Nieberg

Kropat

2016

Proceedings of the 8th International Conference on Agents and Artificial Intelligence

Self Cite

View full text Add to dashboard Cite

show abstract

“…The adaptation is tuned via the learning parameter τ , with smaller values yielding a slower, more robust adaptation and larger values resulting in faster, more efficient, but less robust adaptation. The default value for τ was derived in [8] on the sphere in the limit N → ∞ as τ = 1/ Algorithm 1 (µ/µ I , λ)-σSA-ES 1: g ← 0 2: initialize y (0) , σ (0) ) 3: repeat 4:…”

Section: Test Function and Evolution Strategymentioning

confidence: 99%

“…Equation ( 30) was evaluated on the sphere in [8] in the asymptotic limits N → ∞ (µ ≪ N ) and τ → 0. In this paper a slightly different approach is taken.…”

Section: A Self-adaptation Response (Sar)mentioning

confidence: 99%

“…The left plots, respectively, show the success rate P S ∈ [0, 1] with contour lines showing levels of equal P S (white: P S > 0.9, black: P S < 0.1), see legend at the top. The right plots show log 10 (E r ) ∈ [5,8] displaying the order of magnitude. White regions have no E r -value as P S = 0.…”

Section: Expected Runtime Experimentsmentioning

confidence: 99%

“…Selection by fitness implicitly selects corresponding mutation strengths [6], [7]. σSA-ES have been studied theoretically on the sphere [7], [8], rigde function [8], and the general ellipsoid model [9], which resulted in recommended default parameter settings for the σ-sampling.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Self-Adaptation of Multi-Recombinant Evolution Strategies on the Highly Multimodal Rastrigin Function

Omeradzic,

Beyer

2023

Preprint

View full text Add to dashboard Cite

The self-adaptive, multi-recombinative (µ/µ_I , λ)-ES (Evolution Strategy) is investigated on the highly-multimodal Rastrigin test function by theoretical and experimental means. To this end, the self-adaptation response function is derived in the limit of large populations, which are necessary to achieve high success rates. The established dynamical systems approach is introduced and steady state conditions on Rastrigin are discussed and compared to the sphere function. Then, a characteristic τ is derived to tune the sampling process of the self-adaptive ES. The obtained result is compared to default τ-values. Furthermore, expected runtime experiments are conducted varying τ and population parameters of the ES. Theoretical and experimental results regarding τ are compared in terms of efficiency and robustness showing good agreement.

show abstract

Can Evolution Strategies Benefit from Shrinkage Estimators?

Meyer-Nieberg

Kropat

2018

Transactions on Computational Collective Intelligence XXVIII

View full text Add to dashboard Cite

Sparse Covariance Matrix Adaptation Techniques for Evolution Strategies

Cited by 4 publications

References 11 publications

Evolution Strategies and Covariance Matrix Adaptation - Investigating New Shrinkage Techniques

Evolution Strategies and Covariance Matrix Adaptation - Investigating New Shrinkage Techniques

Self-Adaptation of Multi-Recombinant Evolution Strategies on the Highly Multimodal Rastrigin Function

Can Evolution Strategies Benefit from Shrinkage Estimators?

Contact Info

Product

Resources

About