Abstract:The mutation strength adaptation properties of a multi-recombinative $(\mu/\mu_I, \lambda)$-ES are studied for isotropic mutations. To this end, standard implementations of cumulative step-size adaptation (CSA) and mutative self-adaptation ($\sigma$SA) are investigated experimentally and theoretically by assuming large population sizes ($\mu$) in relation to the search space dimensionality ($N$). The adaptation is characterized in terms of the scale-invariant mutation strength on the sphere in relation to its maximum achievable value for positive progress. %The results show how the different $\sigma$-adaptation variants behave as $\mu$ and $N$ are varied. Standard CSA-variants show notably different adaptation properties and progress rates on the sphere, becoming slower or faster as $\mu$ or $N$ are varied. This is shown by investigating common choices for the cumulation and damping parameters. Standard $\sigma$SA-variants (with default learning parameter settings) can achieve faster adaptation and larger progress rates compared to the CSA. However, it is shown how self-adaptation affects the progress rate levels negatively. Furthermore, differences regarding the adaptation and stability of $\sigma$SA with log-normal and normal mutation sampling are elaborated.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to study the properties of mutation strength adaptation in multi - group Evolution Strategy (ES), especially its performance under large population sizes and in high - dimensional search spaces. Specifically, the author focuses on the adaptation characteristics of Cumulative Step - size Adaptation (CSA) and Mutative Self - Adaptation (σSA) on the sphere function. ### Main problems and objectives 1. **Understanding adaptation characteristics**: Through experimental and theoretical analysis, study the adaptation characteristics of CSA and σSA under different population sizes (µ) and search space dimensions (N). In particular, the author hopes to understand the behavior of these methods under large population sizes and how they affect the performance of the evolution strategy. 2. **Comparing different adaptive methods**: Compare different parameter settings (such as cumulative constants and damping parameters) of standard CSA and σSA, and evaluate their progress rate and convergence speed on the sphere function. 3. **Exploring the impact of adaptive mechanisms**: Explore the negative impact of adaptive mechanisms on the progress rate level, especially the performance differences when using different sampling schemes (such as log - normal distribution and normal distribution). 4. **Laying the foundation for future research**: Through in - depth understanding of these basic adaptation characteristics, provide theoretical support for future research on dynamic population size control to improve the performance of evolution strategies in noisy environments and multimodal optimization problems. ### Research methods - **Experimental and theoretical analysis**: Combine experiments and theoretical derivations to study the performance of CSA and σSA on the sphere function. - **Mathematical modeling**: Establish a mathematical model to derive the progress rate formula of the sphere function and analyze its steady - state characteristics. - **Parametric study**: Study the impact of different parameter settings (such as cumulative constants, damping parameters, etc.) on adaptation characteristics. ### Formula summary - **Sphere function definition**: \[ f(R):=R^{2}, \quad R = \|y\|, \quad y\in\mathbb{R}^{N} \] - **Progress rate definition**: \[ \varphi:=R(g)-\mathbb{E}[R(g + 1)] \] - **Normalized progress rate and mutation intensity**: \[ \varphi^{*}=\frac{\varphi N}{R}, \quad \sigma^{*}=\frac{\sigma N}{R} \] - **Sphere function progress rate formula**: \[ \varphi^{*}=c_{\mu/\mu,\lambda}\sigma^{*}\left(1+\frac{\sigma^{*2}}{2\mu N}\right)\sqrt{1+\frac{\sigma^{*2}}{\mu N}}\left(\sqrt{1+\frac{\sigma^{*2}}{2N}}-N\left(\sqrt{1+\frac{\sigma^{*2}}{\mu N}}-1\right)\right)+O\left(N^{-1/2}\right) \] - **Approximate progress rate formula (under large population sizes)**: \[ \varphi^{*}\approx c_{\vartheta}\sigma^{*}\sqrt{1+\frac{\sigma^{*2}}{2N}}-\frac{\sigma^{*2}}{2\mu} \] Through these

Mutation Strength Adaptation of the $(μ/μ_I, λ)$-ES for Large Population Sizes on the Sphere Function

On the Interaction of Adaptive Population Control with Cumulative Step-Size Adaptation

Improved Running Time Analysis of the (1+1)-ES on the Sphere Function

Towards Self-adaptive Mutation in Evolutionary Multi-Objective Algorithms

A comparison study of self-adaptation in evolution strategies and real-coded genetic algorithms

Insertion-sequence-mediated mutations both promote and constrain evolvability during a long-term experiment with bacteria

Clonal Interference, Multiple Mutations, and Adaptation in Large Asexual Populations

Multidimensional mutations in evolutionary algorithms based on real-valued representation

Self-adjusting Population Sizes for the $(1, λ)$-EA on Monotone Functions

Modulation of Base Specific Mutation and Recombination Rates Enables Functional Adaptation within the Context of the Genetic Code

Some statistical improvements for estimating population size and mutation rate from segregating sites in DNA sequences

Dynamic fitness landscapes: expansions for small mutation rates

A generalization of Kingman's model of selection and mutation and the Lenski experiment

Adaptation through stochastic switching into transient mutators in finite asexual populations

An Empirical Study on the Scalability of Selective Mutation Testing

Adaptation to a changing environment: what me Normal?

Rigorous results for a population model with selection I: evolution of the fitness distribution

Mutation mitigates finite-size effects in spatial evolutionary games

CMA-ES with Two-Point Step-Size Adaptation

Mutation timing in a spatial model of evolution

Entangled Evolution Model: Application to Adaptive Mutation