evolutionary search

This work addresses the inefficiency of traditional genetic algorithms in solving optimization problems due to their reliance on random mutation and recombination, which lack goal-directedness. The authors formulate the problem through the lens of query complexity and propose objective-guided mutation and recombination operators informed by the optimization target. Leveraging reinforcement learning and formal language theory, they analyze the theoretical properties of these operators. For the first time, the study mathematically characterizes the mechanism of goal-directed genetic operators and demonstrates the necessity of population diversity for certain classes of optimization problems. A general model of genetic algorithms is established, enabling the design of a tight algorithm for a specific problem class, and proving that the synergy among generation, mutation, and recombination is essential for efficient optimization.

This study addresses the challenge of protein fitness optimization under low-throughput wet-lab constraints—where only hundreds of sequence-fitness pairs are available. We propose a plug-and-play guidance framework based on discrete diffusion models. Methodologically, we conduct the first systematic evaluation of classifier guidance versus posterior sampling for protein generation and introduce an adaptive guidance strategy inspired by Thompson sampling, replacing data-inefficient reinforcement learning (RL). Crucially, our approach requires no additional pretraining or policy fine-tuning. Under realistic experimental constraints, it significantly outperforms RL baselines, discovering higher-fitness sequences in fewer experimental rounds. Results demonstrate the framework’s practicality, robustness, and plug-and-play efficacy in low-data, low-throughput regimes—enabling efficient, sample-constrained protein engineering without task-specific architectural modifications.

This work proposes a population-based neural combinatorial optimization framework that addresses the limited exploratory capacity and robustness of traditional neural approaches, which typically operate on a single solution. By leveraging neural networks to jointly represent a set of candidate solutions, the framework incorporates a population-aware hierarchical classification mechanism to explicitly model inter-solution information sharing and diversity control. This design simultaneously reinforces high-quality solutions and preserves population diversity, effectively bridging the gap between neural optimization and classical population-based metaheuristics. Experimental results on the Max-Cut and Maximum Independent Set problems demonstrate that the proposed framework substantially improves both solution quality and algorithmic robustness.

This paper investigates the trade-off between solution-set diversity and average quality in black-box optimization: given a fixed budget of solutions, maximize their average fitness while ensuring that the pairwise distance between any two solutions exceeds a predefined threshold. We propose the first systematic empirical framework to quantify the performance limits of mainstream heuristic algorithms—including evolutionary algorithms and random search—on this task, and analyze their dependence on problem characteristics. A key finding is that uniform random sampling (RS) significantly outperforms trajectory-based heuristics across the vast majority of benchmark problems, establishing an unexpected yet robust strong baseline. This result challenges prevailing algorithmic design paradigms and provides critical empirical evidence—and a new research direction—for developing algorithms that simultaneously generate high-quality and diverse solution sets.

Quality-Diversity (QD) methods remain underexplored and theoretically underdeveloped for dynamic combinatorial optimization problems, such as dynamic scheduling. Method: This paper proposes the first QD genetic programming framework tailored for dynamic environments. It introduces a novel behavior-representation mapping mechanism that projects heuristic-rule genotypes into a measurable behavior space, enabling the construction of a QD map that jointly optimizes solution quality and behavioral diversity. Integrated with dynamic instance training and adaptive map updating, the framework supports continual evolution of robust scheduling heuristics. Results: Experiments demonstrate that our approach significantly outperforms baselines on both static and dynamic scheduling benchmarks. It successfully generates high-quality, behaviorally diverse heuristic sets and—critically—reveals, for the first time, the evolutionary dynamics of QD maps under environmental non-stationarity.

This work addresses the limitation of conventional survival selection in evolutionary diversity optimization, which often fails due to its dependence on pairwise solution diversity. To overcome this issue, we propose a novel framework that enables the synchronous generation of multiple candidate solutions per generation, along with a tailored survival selection mechanism designed specifically for this setting. By moving beyond the traditional paradigm of single-solution, sequential updates, our approach effectively handles the dynamic nature of each solution’s contribution to population diversity. Experimental results demonstrate that, under certain conditions, the proposed multi-solution generation strategy accelerates convergence toward diverse solutions and significantly improves both the spread and quality balance of the final solution set.

This work addresses the challenges of high computational cost, large variance, and performance instability under limited evaluation budgets in large language model (LLM)-driven program evolution. To overcome these issues, the authors propose TurboEvolve, a multi-island evolutionary framework that innovatively integrates LLM-based multi-candidate verbal sampling, an adaptive online scheduling mechanism, and a clustering-based seed pool injection strategy. This approach enhances elite preservation while maintaining population diversity. TurboEvolve significantly improves sample efficiency and evolutionary robustness, outperforming state-of-the-art methods on multiple program optimization benchmarks with fewer evaluations and establishing new best-known solutions for several tasks.

This work addresses key limitations in existing large language model (LLM)-based evolutionary search, which struggles to distinguish semantically equivalent yet syntactically diverse programs and fails to effectively preserve strategic potential or detect saturation within strategy families. To overcome these challenges, the authors propose introducing a strategy-space layer into program evolution, elevating natural language strategy descriptions to first-class evolutionary states. By integrating strategy representation, hierarchical experience retrieval, and strategy-landscape navigation, the method enables explicit organization, memory, and exploration at the strategy level. Combining LLMs with evolutionary algorithms, strategy clustering, behavior-complementarity-based retrieval, and summarization mechanisms, the approach significantly outperforms baseline methods across tasks in mathematical algorithm discovery, system optimization, and agent framework design, achieving a relative improvement of 21% in open-ended system optimization.

This work proposes an efficient framework for automatic algorithm design that decouples algorithm discovery from costly real-world evaluations, addressing the limitations of existing large language model–based approaches which require extensive and expensive assessments on actual problems. By leveraging surrogate functions to enable deep exploration in the algorithm space and incorporating problem landscape features to guide the co-evolution of genetic programming and large language models, the method generates optimization algorithms with strong generalization capabilities. Experimental results demonstrate that the proposed approach significantly reduces the number of required real evaluations while successfully discovering high-performance algorithms across multiple practical optimization tasks, thereby validating its effectiveness and practicality in resource-constrained settings.

Current large language model (LLM)-driven approaches to automated heuristic design struggle to effectively solve complex optimization problems characterized by tightly coupled subproblems. This work proposes CoupleEvo, a novel framework that extends LLM-guided heuristic evolution to coupled optimization scenarios for the first time. It introduces three cooperative evolution strategies—sequential, iterative, and integrated—and systematically investigates how different coordination mechanisms affect search stability and solution quality. Experimental results on two representative classes of coupled optimization problems demonstrate that decomposition-based strategies (sequential and iterative) significantly outperform the integrated approach, exhibiting both more stable convergence behavior and superior solution quality.