LLM-Based Method for Discovering Differential Equations from Data
A new technique named DoLQ has been developed by researchers to extract ordinary differential equations from observational data using large language models (LLMs). This method overcomes a significant drawback of current symbolic regression methods, which often focus on quantitative metrics while overlooking essential domain knowledge for physical validity. DoLQ utilizes a multi-agent system: a Sampler Agent generates potential dynamic systems, a Parameter Optimizer enhances the equations for precision, and a Scientist Agent employs an LLM for both qualitative and quantitative assessments, iteratively refining the search based on the findings. Tests on multi-dimensional ordinary differential equation benchmarks indicate that DoLQ surpasses existing approaches. The findings are published in arXiv preprint 2605.07323.
Key facts
- DoLQ uses LLMs for qualitative and quantitative evaluation in equation discovery.
- The method employs a multi-agent architecture with Sampler, Parameter Optimizer, and Scientist Agents.
- It addresses the gap of incorporating domain knowledge for physical plausibility.
- Experiments on multi-dimensional ODE benchmarks show superior performance.
- The research is published on arXiv with ID 2605.07323.
- Existing symbolic regression methods rely primarily on quantitative metrics.
- DoLQ iteratively guides the search for governing differential equations.
- The Scientist Agent leverages an LLM to synthesize evaluation results.
Entities
Institutions
- arXiv