Formulating AutoML as a Variable-Length Optimization Problem: A Tree of Thought Approach with LLM-Driven Code Generation
Keywords: AutoML, Tree of Thought, LLM
Abstract: Recent advancements in machine learning have created a demand for automated systems that enable efficient development and deployment of machine learning applications. Traditional Automated Machine Learning (AutoML) approaches often rely on fixed pipeline structures, which limit adaptability to diverse task complexities. In this paper, we introduce a novel formulation of AutoML as a variable-length optimization problem, allowing for dynamic adjustment of model architectures based on task requirements. To effectively navigate the expanded search space of variable-length models, we employ the Tree of Thoughts (ToT) method combined with Large Language Models (LLMs). This framework utilizes a sequential decision-making process, allowing models to be incrementally constructed by evaluating prior outcomes. Additionally, LLMs automatically generate the code corresponding to each decision, transforming model configurations into executable pipelines and reducing manual intervention. Our approach enhances efficiency by focusing on promising pathways and improves transparency by explicitly showcasing how each decision contributes to the overall optimization. Experiments conducted on diverse datasets, including OpenML and clinical tasks, demonstrate that our method outperforms traditional AutoML systems, delivering superior model performance and better adaptability across different task complexities.
Primary Area: optimization
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 7005
Loading