Enhancing Large Language Models-Based Code Generation by Leveraging Genetic Improvement

Created by W.Langdon from gp-bibliography.bib Revision:1.8051

@InProceedings{Pinna:2024:EuroGP,

• author = "Giovanni Pinna and Damiano Ravalico and Luigi Rovito and Luca Manzoni and Andrea {De Lorenzo}",
• editor = "Mario Giacobini and Bing Xue and Luca Manzoni",
• title = "Enhancing Large Language Models-Based Code Generation by Leveraging Genetic Improvement",
• booktitle = "EuroGP 2024: Proceedings of the 27th European Conference on Genetic Programming",
• year = "2024",
• volume = "14631",
• series = "LNCS",
• pages = "108--124",
• publisher = "Springer",
• address = "Aberystwyth",
• month = "3-5 " # apr,
• organisation = "EvoStar, Species",
• keywords = "genetic algorithms, genetic programming, Genetic Improvement, Grammatical Evolution, Evolutionary Computation, Evolutionary Algorithms, Large Language Models, LLM, Artificial Intelligence, AI, Machine Learning, Neural Networks, ANN, Code Generation",
• isbn13 = "978-3-031-56957-9",
• URL = "https://arts.units.it/handle/11368/3071899",
• DOI = "doi:10.1007/978-3-031-56957-9_7",
• code_url = "https://github.com/dravalico/LLMGIpy",
• size = "17 pages",
• abstract = "In recent years, the rapid advances in neural networks for Natural Language Processing (NLP) have led to the development of Large Language Models (LLMs), able to substantially improve the state-of-the-art in many NLP tasks, such as question answering and text summarisation. Among them, one particularly interesting application is automatic code generation based only on the problem description. However, it has been shown that even the most effective LLMs available often fail to produce correct code. To address this issue, we propose an evolutionary-based approach using Genetic Improvement (GI) to improve the code generated by an LLM using a collection of user-provided test cases. Specifically, we employ Grammatical Evolution (GE) using a grammar that we automatically specialize, starting from a general one, for the output of the LLM. We test 25 different problems and 5 different LLMs, showing that the proposed method is able to improve in a statistically significant way the code generated by LLMs. This is a first step in showing that the combination of LLMs and evolutionary techniques can be a fruitful avenue of research.",
• notes = "Part of \cite{Giacobini:2024:GP} EuroGP'2024 held in conjunction with EvoCOP2024, EvoMusArt2024 and EvoApplications2024",

}

Genetic Programming entries for Giovanni Pinna Damiano Ravalico Luigi Rovito Luca Manzoni Andrea De Lorenzo

Citations