13 papers · ranked by Valyu relevance
Leyi Zhao, Weijie Huang, Yitong Guo, Jiang Bian + 2 more
Optimizing scientific computing algorithms for modern GPUs remains a labor-intensive and iterative process, requiring repeated cycles of code modification, benchmarking, and tuning across complex hardware and software stacks. While recent work has explored large language model (LLM)–assisted evolutionary algorithms for…
Gang Liu, Yihan Zhu, Jie Chen, Meng Jiang
Large language models hold promise as scientific assistants, yet existing agents either rely solely on algorithm evolution or on deep research in isolation, both of which face critical limitations. Pure algorithm evolution, as in AlphaEvolve, depends only on the internal knowledge of LLMs and quickly plateaus in…
Sartori, Camilo Chacón, Blum, Christian
Automatic algorithm configuration tools such as irace efficiently tune parameter values but leave algorithmic code unchanged. This paper introduces a first version of irace-evo, an extension of irace that integrates code evolution through large language models (LLMs) to jointly explore parameter and code spaces. The…
Minhua Lin, Hanqing Lu, Zhan Shi, Bing He + 10 more
As Large Language Models (LLMs) move from curated training sets into open-ended real-world environments, a fundamental limitation emerges: static training cannot keep pace with continual deployment environment change. Scaling trainingtime and inference-time compute improves static capability but does not close this…
Kaichen Ouyang, Yu, Mingyang, Ke + 11 more
This paper introduces a novel framework linking evolutionary computation to statistical physics by formulating optimization as a statistical phase transition. We propose Wasserstein Evolution (WE), an algorithm based on the Wasserstein gradient flow of a free energy functional, translating the physical competition…
Rongjie Liao, Junhao Qiu, Xin Chen, Xiaoping Li
Customized static operator design has enabled widespread application of Evolutionary Algorithms (EAs), but their search performance is transient during iterations and prone to degradation. Dynamic operators aim to address this but typically rely on predefined designs and localized parameter control during the search…
Zongmin Yu, Liu Yang
We introduce Evolutionary Ensemble (EvE), a decentralized framework that organizes existing, highly capable coding agents into a live, co-evolving system for algorithmic discovery. Rather than reinventing the wheel within the "LLMs as optimizers" paradigm, EvE fixes the base agent substrate and focuses entirely on…
Jiayi Zhang, Yongfeng Gu, Jianhao Ruan, Maojia Song + 9 more
Agentic evolution has emerged as a powerful paradigm for improving programs, workflows, and scientific solutions by iteratively generating candidates, evaluating them, and using feedback to guide future search. However, existing methods are typically instantiated either as fixed hand-designed procedures that are…
Inès Benito, Johannes F. Lutzeyer, Benjamin Doerr
Baldwinian and Lamarckian evolution have existed for a long time in evolutionary algorithms (EAs) without ever dominating the academic literature or practical applications. In this work, we use modern empirical and theoretical methods to revisit Lamarckian and Baldwinian evolution and rigorously compare them with the…
Minghao Yan, Bo Peng, Benjamin Coleman, Ziqi Chen + 10 more
Large language models have become drivers of evolutionary search, but most systems rely on a fixed, prompt-elicited policy to sample next candidates. This limits adaptation in practical engineering and research tasks, where evaluations are expensive, and progress depends on learning task-specific search dynamics. We…
Yoichi Ishibashi, Taro Yano, Masafumi Oyamada
AlphaEvolve and FunSearch have demonstrated the potential of combining large language models (LLMs) with evolutionary search for automated algorithm discovery. However, discovery success is shaped not only by model capability but also significantly by the design of the execution infrastructure, i.e., the harness. This…
Nam Hai Le
The gene-centric paradigm, formalized in the Modern Synthesis and operationalized through genetic algorithms, attributes evolutionary causality exclusively to genes; organisms are passive vehicles for genetic replication. Denis Noble's phenotype-first framework challenges this view, arguing that organisms are active…
Fabian Hoppe, Melven Röhrig-Zöllner, Philipp Knechtges
We consider LLM-based algorithm development through a case study on contractionorder optimisation for tensor networks with OpenEvolve. We pay particular attention to the choice of the LLM as well as design choices such as evaluation metric and test instances. Our results highlight both the promise of verifier-guided…