MOT-SR: Multi-Objective Tool-Augmented Symbolic Regression with LLMs
A novel approach named Multi-Objective Tool-augmented Symbolic Regression (MOT-SR) has been introduced to enhance the process of discovering equations from observational data. This technique incorporates external analytical tools to derive structural priors that assist in the generation of equations, while also optimizing for accuracy, complexity, and generalization through a multi-objective evaluation framework. This development addresses the shortcomings of current Large Language Model (LLM) methods, which do not incorporate data analysis for identifying variable dependencies and are limited to a single-objective evaluation that focuses only on fitting error. The research paper can be found on arXiv under ID 2607.29561.
Key facts
- MOT-SR is a unified framework for symbolic regression.
- It integrates external analytical tools to extract structural priors.
- It jointly optimizes accuracy, complexity, and generalization.
- It addresses limitations of LLM-based approaches: lack of data analysis and single-objective evaluation.
- The paper is available on arXiv with ID 2607.29561.
- The framework aims to improve efficiency and exploration of equation space.
Entities
Institutions
- arXiv