RESEARCH · RESEARCH · #2103
L2G-SR: learned structural verifier guides LLM-based symbolic regression
L2G-SR is a framework that augments LLM-based symbolic regression with a learned structural verifier trained using tree edit distance (TED) to score formula structural similarity; at search time the verifier complements numerical error for candidate selection and a generator refines expressions via analyst-guided revisions. On LLM-SRBench, L2G-SR achieves state-of-the-art numerical fit, equation-level accuracy, and TED across evaluated LLM backbones and domains.
KEY POINTS
- L2G-SR is a framework that augments LLM-based symbolic regression with a learned structural verifier trained using tree edit distance (TED) to score formula structural similarity; at search time the verifier complements numerical error for candidate selection and a generator refines expressions via analyst-guided revisions.
- On LLM-SRBench, L2G-SR achieves state-of-the-art numerical fit, equation-level accuracy, and TED across evaluated LLM backbones and domains.
- By learning to evaluate structural similarity (via TED) and combining that with numerical error, L2G-SR improves equation recovery and highlights the value of structural guidance for LLM-driven symbolic regression.
WHY IT MATTERS
By learning to evaluate structural similarity (via TED) and combining that with numerical error, L2G-SR improves equation recovery and highlights the value of structural guidance for LLM-driven symbolic regression.