Jul 2026· GECCO Companion· pp. 7-8· 0 citations· 7 references
Computer Science
TL;DR
The results suggest that recent progress in language models and tool use may already be sufficient to support practical automated metaheuristic design, and that recent progress in language models and tool use may already be sufficient to support practical automated metaheuristic design.
Abstract
This paper presents our submission to the competition on LLM-designed evolutionary algorithms for the GNBG benchmark suite. We study a simple alternative to recent automated algorithm-design frameworks: instead of placing the language model inside a specialized evolutionary loop, we treat optimizer construction as an ordinary coding task. In our setup, a general-purpose coding agent iteratively modifies and evaluates a candidate optimizer inside a small repository that provides a fixed interface, benchmarking harness, and lightweight experiment log. The main question is whether such an agent, without any bespoke search controller, can still discover competitive metaheuristics. Our results indicate that the answer is positive: the resulting optimizer performs well on a substantial subset of GNBG II. These findings suggest that recent progress in language models and tool use may already be sufficient to support practical automated metaheuristic design.
We propose a framework for automated discovery and composition of metaheuristic operators using large language models (LLMs). The method begins with generation of candidate algorithms, followed by LLM-guided selection of impactful operators, iterative construction of specialized operator variants, and final synthesis of an adaptive optimizer. The approach emphasizes task coverage rather than single-metric optimization, producing complementary operators specialized for different problem characteristics. These operators are subsequently integrated into an adaptive algorithm via LLM-driven parameter control and selection mechanisms. This work is submitted as a competition entry for LLM-generated metaheuristics evaluated on the GNBG benchmark.
Paweł Kolendo, Michal Pluháček· GECCO Companion· 0 citations
This paper presents an approach to automate the evolution of mutation operators in Hybrid Genetic Search (HGS) for the Capacitated Vehicle Routing Problem (CVRP) using Large Language Models (LLMs). Despite the advanced nature of HGS, its mutation heuristics are typically hand-crafted. Recent studies have demonstrated the capacity of LLMs to produce effective routing heuristics; however, these endeavours have predominantly overlooked the utilisation of mutation operators in favour of crossover operators. To bridge this gap, we integrate LLMs within the ruin-and-re create framework of HGS. This work establishes a pathway toward automated fine-grained optimisation of metaheuristic components.
This paper presents the competition entry on LLM-designed evolutionary algorithms for the GNBG-generated benchmark suite. We use EASE (Effortless Algorithmic Solution Evolution), a modular framework that prompts a large language model to generate complete optimizers, evaluates them under the competition protocol, and feeds performance summaries together with improvement-oriented analysis back into subsequent iterations. The automated process produced ten valid candidate algorithms. The best generated solver was a hybrid adaptive Differential Evolution method combining L-SHADE-style parameter adaptation, archive-based diversity preservation, covariance-inspired variation, and budget-aware local refinement. The results show that iterative LLM-guided refinement can produce effective, benchmark-specific optimizers.
T. Kadavy, Jozef Kovác, Adam Viktorin et al.· GECCO Companion· 0 citations
Comprehensive experiments on numerical benchmarks and neuroevolution robotic control tasks demonstrate that AutoPSO consistently discovers novel PSO variants that significantly outperform strong baselines and confirm that AutoPSO achieves increasing performance gains with larger swarm sizes.
Xin-Meng Yu, Jia-Xin Gao, Jianguo Zhang et al.· IEEE Transactions on Evoluti...· 0 citations
Evolutionary algorithms typically follow very rigid generational cycles. These cycles are simple to implement, analyze, and use, but they pose severe constraints on design space. Consequently, typical implementations can be differentiated only across a limited range of parameters and paradigms. While parameter tuning and human expertise can help determine which algorithmic configuration may be best on a given problem class, it is likely that there exist even better optimizers outside of this traditional design space. Previous work has shown that directed graphs can be used to automatically design evolutionary cycles which outperform traditional configurations on specific benchmarks. In this work, we present a novel representation and associated variation operator for evolving such graphs inspired by techniques used for neuro-evolution, demonstrate that this can significantly improve performance and result in more complex structures, and show evidence suggesting that, for some problems, traditional cycles are not even locally optimal. We further validate the effectiveness of our algorithm against benchmarks from the IOHprofiler suite, achieving competitive results.
Braden N. Tisdale, D. Tauritz, A. Pope· Annual Conference on Genetic...· 0 citations
SAGE, a modular approach that trains a small searcher model to handle codebase exploration as a tool for a frozen large solver model, is proposed, demonstrating a practical path to cheaper, modular software engineering agents.
N. Sorokin, I. Sedykh, Timur Ionov et al.· Annual International ACM SIG...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
