A Comparison and Analysis of LLM Agents within the Context of An LLM-Guided Evolution for Object Detection

Created by W.Langdon from gp-bibliography.bib Revision:1.8519

@InProceedings{yu:2025:GECCOcomp,

• author = "YiMing Yu and Jason Zutty",
• title = "A Comparison and Analysis of {LLM} Agents within the Context of An {LLM-Guided} Evolution for Object Detection",
• booktitle = "Proceedings of the 2025 Genetic and Evolutionary Computation Conference Companion",
• year = "2025",
• editor = "Ryan Urbanowicz and Will N. Browne",
• pages = "327--330",
• address = "Malaga, Spain",
• series = "GECCO '25 Companion",
• month = "14-18 " # jul,
• organisation = "SIGEVO",
• publisher = "Association for Computing Machinery",
• publisher_address = "New York, NY, USA",
• keywords = "genetic algorithms, genetic programming, grammatical evolution, computer aided/automated design, automated machine learning, large language models, neuroevolution, Evolutionary Machine Learning: Poster",
• isbn13 = "979-8-4007-1464-1",
• URL = "https://doi.org/10.1145/3712255.3726669",
• DOI = "doi:10.1145/3712255.3726669",
• size = "4 pages",
• abstract = "In machine learning, evolutionary algorithms have traditionally depended on fixed rules and predefined building blocks. The Guided Evolution (GE) framework transforms this approach by incorporating Large Language Models (LLMs) to directly modify code and intelligently guide both mutations and crossovers. A key element of GE is the {"}Evolution of Thought{"} (EoT) technique, which establishes positive feedback loops, allowing LLMs to refine their decisions iteratively based on successful results. Building on this concept, we introduced the LLM-guided evolution to modify the architecture of a You Only Look Once (YOLO) model and enhance its performance on the KITTI dataset. Our findings show that LLM-Guided Evolution produced variants with significant performance improvements, such as an increase in Mean Average Precision from 92.5\% to 94.7\%. we also evaluated various LLM-GEs with different LLMs, including Mixtral-8x7B, Llama-3.1-70B, and Llama-3.3-70B, to assess their impacts on the evolutionary process. Our analysis shows that GE with Llama-3.1 and Llama-3.3 surpassed Mixtral-8x7B in performance, underscoring the importance of LLMs selection in refining evolutionary algorithms. These results highlight the flexibility and effectiveness of GE in real-world computer vision challenges, offering a novel paradigm for autonomous model optimization that combines LLM-driven reasoning with evolutionary strategies.",
• notes = "GECCO-2025 EML A Recombination of the 34th International Conference on Genetic Algorithms (ICGA) and the 30th Annual Genetic Programming Conference (GP)",

}

Genetic Programming entries for YiMing Yu Jason Zutty

Citations