Abstract:Three state-of-the-art adaptive population control strategies (PCS) are theoretically and empirically investigated for a multi-recombinative, cumulative step-size adaptation Evolution Strategy $(\mu/\mu_I, \lambda)$-CSA-ES. First, scaling properties for the generation number and mutation strength rescaling are derived on the sphere in the limit of large population sizes. Then, the adaptation properties of three standard CSA-variants are studied as a function of the population size and dimensionality, and compared to the predicted scaling results. Thereafter, three PCS are implemented along the CSA-ES and studied on a test bed of sphere, random, and Rastrigin functions. The CSA-adaptation properties significantly influence the performance of the PCS, which is shown in more detail. Given the test bed, well-performing parameter sets (in terms of scaling, efficiency, and success rate) for both the CSA- and PCS-subroutines are identified.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **How to optimize the interaction between the self - adaptive population control strategy (PCS) and the cumulative step - size adaptation (CSA) evolution strategy (ES) to improve the optimization performance on noisy and multimodal functions.** Specifically, the paper studied three state - of - the - art self - adaptive population control strategies (PCS) and applied them to the multi - recombinative cumulative step - size adaptation evolution strategy ((µ/µI, λ)-CSA - ES). The research focuses include: 1. **Scaling property analysis**: The scaling properties of the rescaling of generations and mutation intensities under large population sizes were derived, especially their performance on the sphere function. 2. **Adaptability study of CSA variants**: The adaptability of three standard CSA variants was studied, and their changes with population size and dimension were analyzed. 3. **Experimental verification**: The effects of different combinations of PCS and CSA were verified by testing on the sphere function, random function, and Rastrigin function. 4. **Parameter selection**: Based on the test results, parameter sets with good performance in terms of scaling, efficiency, and success rate were identified. ### Core problem of the paper The core problem of the paper is to explore how, on different test functions (such as the sphere function, random function, and Rastrigin function), by adjusting the population size and mutation intensity, the self - adaptive population control strategy (PCS) and the cumulative step - size adaptation (CSA) evolution strategy (ES) can work better together, thereby improving the optimization performance. ### Key points - **Self - adaptive population control strategy (PCS)**: It can dynamically adjust the population size according to the algorithm performance to deal with strong noise and highly multimodal problems. - **Cumulative step - size adaptation (CSA)**: It is used to adjust the global mutation intensity to ensure an appropriate balance between exploration and exploitation during the optimization process. - **Combination of theory and experiment**: Through theoretical derivation and experimental verification, the influence of different CSA variants on the performance of PCS was analyzed, and optimization suggestions were proposed. ### Formula summary Some of the key formulas involved in the paper are as follows: - Definition of progress rate: \[ \varphi(g) := R(g) - E[R(g + 1)] \] - Normalized progress rate: \[ \varphi^*=\frac{\varphi N}{R}, \quad \sigma^*=\frac{\sigma N}{R} \] - Generation calculation formula: \[ G=\frac{N}{\varphi^*}\ln\left(\frac{R(g_0)}{R(g)}\right) \] - Steady - state mutation intensity: \[ \sigma_{ss}^*=\gamma\sigma_{\varphi_0}^*=\gamma(8N)^{1/4}(c_\vartheta\mu)^{1/2} \] These formulas help to understand the behavior of CSA - ES under large population sizes and provide a theoretical basis for the selection and adjustment of PCS.

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

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

A dimensional difference-based population size adjustment framework for differential evolution

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

Lexicase Selection Parameter Analysis: Varying Population Size and Test Case Redundancy with Diagnostic Metrics

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

Covariance matrix adaptation evolution strategy based on correlated evolution paths with application to reinforcement learning

CMA-ES with Two-Point Step-Size Adaptation

Adaptive multi-context cooperatively coevolving particle swarm optimization for large-scale problems

Adaptation of the Scaling Factor Based on the Success Rate in Differential Evolution

Parameter and strategy adaptive differential evolution algorithm based on accompanying evolution

Evolutionary game dynamics of combining two different aspiration-driven update rules in structured populations

Optimised control of adaptive evolution with competing selective pressures

Improving genetic algorithms performance via deterministic population shrinkage

Large-Scale Evolution Strategy Based on Search Direction Adaptation

An Adaptive Framework to Tune the Coordinate Systems in Evolutionary Algorithms

Differential evolution algorithm with multi-population cooperation and multi-strategy integration

Mobility-Dependent Selection of Competing Strategy Associations

Surrogate Model Assisted Cooperative Coevolution for Large Scale Optimization

Population reduction with individual similarity for differential evolution

A Multi-Strategy Differential Evolution Algorithm with Adaptive Similarity Selection Rule