A Longitudinal Analysis of the CEC Single-Objective Competitions (2010–2024) and Implications for Variational Quantum Optimization

Vojtěch Novák, Tomáš Bezděk, Ivan Zelinka, Swagatam Das, Martin Beseda · 2026

Read A Longitudinal Analysis of the CEC Single-Objective Competitions (2010–2024) and Implications for Variational Quantum Optimization by Vojtěch Novák, Tomáš Bezděk, Ivan Zelinka, Swagatam Das, Martin Beseda (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-24140.txt.

Introduction

Novák and co-authors review fifteen years of the CEC single-objective optimization contests, from 2010 to 2024. They study how benchmark design shaped the winning algorithms. Dense rotation matrices, added in 2014, acted as a strong filter. They hurt coordinate-based methods and lifted Differential Evolution variants such as L-SHADE. After 2020, winners grew into complex hybrids that mix mechanisms for stable ranks. The authors see parallels with variational quantum algorithm landscapes. Evolved CEC solvers may hold skills that quantum control needs, they suggest.

Catalog record

CREATORVojtěch Novák, Tomáš Bezděk, Ivan Zelinka, Swagatam Das, Martin Beseda
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE92613
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.24140

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A Longitudinal Analysis of the CEC Single-Objective Competitions (2010–2024) and Implications for Variational Quantum Optimization Vojtěch Novák, Tomáš Bezděk, Ivan Zelinka, Swagatam Das, Martin Beseda Abstract This paper provides a historical analysis of the IEEE CEC Single Objective Optimization competition results (2010–2024). We analyze how benchmark functions shaped winning algorithms, identifying the 2014 introduction of dense rotation matrices as a key performance filter. This design choice introduced parameter non-separability, reduced effectiveness of coordinate-dependent methods (PSO, GA), and established the dominance of Differential Evolution variants capable of preserving the rotational invariance of their difference vectors, specifically L-SHADE. Post-2020 analysis reveals a shift towards high complexity hybrid optimizers that combine different mechanisms (e.g., Eigenvector Crossover, Societal Sharing, Reinforcement Learning) to maximize ranking stability. We conclude by identifying structural similarities between these modern benchmarks and Variational Quantum Algorithm landscapes, suggesting that evolved CEC solvers possess the specific adaptive capabilities required for quantum control. 1 Introduction Global optimization of continuous functions is fundamental to computational intelligence and is essential for solving complex problems in engineering and science. Since their inception, competitions organized within the IEEE Congress on Evolutionary Computation (CEC) have established themselves as the de facto standard for benchmarking stochastic optimization algorithms [ 131 ] . These annual events provide a rigorous standardized framework, defined by their respective problem definition reports [ 187 , 50 , 111 , 112 , 209 , 129 , 100 ] . This allows researchers to evaluate the performance of novel algorithms against state-of-the-art methods under controlled experimental conditions. The resulting datasets and rankings, which we collated from the official repositories maintained by the competition organizers [ 177 ] , have significantly influenced the trajectory of evolutionary computation driving the transition from static canonical algorithms to highly adaptive and complex hybrid systems. While annual reports summarize specific competition results, broader longitudinal analyses are rare. Previous studies by Molina et al. [ 133 ] and Skvorc et al. [ 167 ] have critically examined whether newer winners statistically outperform their predecessors, often finding that progress is sometimes the result of overfitting to a specific year’s suite rather than general algorithmic improvement. Existing surveys typically focus on specific benchmark suites or provide a snapshot of performance for a single year, often comparing a proposed algorithm only against its immediate predecessors. However, these studies are limited to specific timeframes and provide fragmented views of results. This paper complements those statistical surveys by analyzing the design evolution of the winning algorithms from 2010 to 2024. Rather than simply tabulating rankings, we trace the lineage of dominant mechanisms, such as Success History Adaptation [ 184 ] and Linear Population Size Reduction [ 181 ] , to understand how solvers adapted to specific changes in benchmark topology. For example, we identify how the introduction of dense rotation matrices in the CEC 2014 hindered the performance of coordinate-dependent methods such as Particle Swarm Optimization (PSO) [ 94 ] and Genetic Algorithms (GA) [ 71 ] . This shift forced the community toward Differential Evolution [ 51 ] variants that leverage rotationally invariant mutation [ 180 , 182 , 146 ] , where the addition of population reduction proved decisive in outperforming restart based strategies like Covariance Matrix Adaptation Evolution Strategy (CMA-ES). The analysis identifies three distinct evolutionary eras that characterize the progress of the field during this period. The first phase spanning from 2010 to 2013 represents a period of algorithmic specialization, where diverse paradigms competed and Canonical Genetic Algorithms claimed victory on Real-World Problems [ 59 ] . The second phase, beginning in 2014, marks the dominance of adaptive Differential Evolution with the rise of the L-SHADE family [ 181 , 11 , 34 ] , establishing the superiority of adaptation over static heuristics. By 2017, this evolved into a period of hybridization as benchmark functions became increasingly non-separable and multimodal through the use of composition functions. This led to the proliferation of societal algorithms that structurally hybridized distinct search logics to navigate composite landscapes such as AGSK [ 128 ] and IMODE [ 159 ] . The third and most recent phase from 2021 to 2024 reflects a period of intense complexity and refinement characterized by the adoption of Eigenvector Crossover to mimic covariance learning [ 37 , 4 ] and Success Rate based adaptation to sharpen exploitation [ 173 ] . We examine the post 2020 phase and its increasing complexity of algorithms, noting the tendency to combine complexity with necessary innovation [ 24 ] . We extend our analysis of CEC results into a forward-looking proposal to apply these evolved optimizers to Variational Quantum Algorithms (VQAs) [ 43 , 124 , 93 , 84 , 21 ] . VQAs leverage hybrid quantum-classical systems to enhance machine learning models with quantum features [ 7 ] , solve large-scale combinatorial optimization problems, and simulate entangled quantum systems. However, their parameter landscapes feature extreme noise, entangled variable interactions, and barren plateaus [ 104 ] , challenges akin to post-2014 CEC benchmarks’ non-separability and multimodality. Researchers have already explored gradient-free evolutionary methods to address these issues, such as basic DE to avoid local minima [ 65 ] and mitigate barren plateaus [ 7 ] , as well as comparisons of standard DE and CMA-ES variants on VQA performance [ 31 ] . While these foundational applications demonstrate the viability of evolutionary approaches, they primarily rely on canonical or basic implementations, leaving room for enhancement through more adaptive mechanisms. CEC-dominant methods, like L-SHADE variants, are particularly suited to extend this work due to their rotational invariance for correlated parameters due to quantum entanglement. The next biggest challenge is finite sampling noise of the quantum state that can be approached with, for example, population averaging in CMA-ES [ 9 , 81 , 135 ] . 2 Methodology This study employs a systematic analysis of the winning algorithms from the CEC Single Objective Optimization competitions from 2010 through 2024. While the raw ranking data was collated from the official repositories maintained by the organizers [ 177 ] , our analytical framework moves beyond simple tabulation to categorize the evolutionary mechanisms that enabled these algorithms to survive increasing benchmark complexity. The primary data sources for this review are the technical reports and algorithmic descriptions of the top-performing entries (Ranks 1–3) spanning fifteen years of competition. Specific algorithmic details, operator definitions, and parameter adaptation strategies were extracted directly from the individual algorithm papers summarized in the tables in Section 3 . To accurately map these evolutionary trajectories, we ground our classification in the canonical definitions of the major algorithm families. We analyze foundational frameworks established by Holland for Genetic Algorithms [ 82 ] , Storn and Price for Differential Evolution [ 176 ] , and Hansen and Ostermeier for Evolution Strategies [ 79 ] . This theoretical baseline allows us to identify the specific structural limitations that led to the decline of coordinate-dependent methods like PSO [ 162 ] and the parallel rise of rotationally invariant strategies [ 180 ] . Our methodology is based on the foundational literature in the field. Das and Suganthan established the baseline taxonomy for Differential Evolution in their surveys [ 51 , 49 ] , yet their analysis predates the widespread adoption of the

Read the complete text