Research map / Large language models / Agents, code and math
Robust Equation Structure learning with Adaptive Refinement
Yunlun Li, Sinno Jialin Pan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add initialization (ablation) | RESTART · BPG12 (ID and OOD) | NMSE | Improved |
| add short-term guidance (boosting analysis) (ablation) | RESTART · BPG12 (ID and OOD) | NMSE | Improved |
| add long-term structure memory (ablation) | RESTART · BPG12 (ID and OOD) | NMSE | Improved |
| use structural boosting equation instead of additive residual fit (ablation) | RESTART · BPG12 (ID and OOD) | NMSE | Improved |
| use RESTART instead of LLMSR with Qwen3-8B backbone | LLMSR · BPG18 (ID) | NMSE: 0.20248 -> 0.02254 | Improved |
| use RESTART instead of LLMSR with Qwen3-8B-think backbone | LLMSR · BPG18 (ID) | NMSE: 0.90352 -> 0.00801 | Improved |
| use RESTART instead of LLMSR with Qwen3-1B backbone | LLMSR · BPG18 (OOD) | NMSE: 0.69452 -> 0.77328 | Hurt |
| swap E2E solver for PySR solver (ablation) | RESTART · BPG18 (ID) | NMSE: 0.02254 -> 0.00122 | Improved |
| swap E2E solver for polynomial regression solver (ablation) | RESTART · BPG18 (ID) | NMSE: 0.02254 -> 0.47034 | Hurt |
| use only 25% of the iterations | RESTART · four hard Biology problems | NMSE | No clear effect |
| use RESTART instead of LLMSR on Material Science OOD | LLMSR · LSR-Synth Material Science (OOD) | Acc0.1: 96.00 -> 100.00 | Improved |
| use RESTART instead of LLMSR on Physics in-distribution | LLMSR · LSR-Synth Physics (ID) | NMSE: 0.002 -> 0.003 | No clear effect |
Closest papers by meaning
- FunctionEvolve: Structure-Guided Symbolic Regression with LLMs (NeurIPS 2026 · similarity 0.90)
- LLM-SR: Scientific Equation Discovery via Programming with Large Language Models (ICLR 2025 · similarity 0.88)
- SMILE: Bridging Continuous Optimization and Discrete Symbolic Recovery (NeurIPS 2026 · similarity 0.88)
- Mimicking the Physicist's Eye: A VLM-centric Approach for Physics Formula Discovery (NeurIPS 2026 · similarity 0.87)
- Post Hoc Regression Refinement via Pairwise Rankings (NeurIPS 2025 · similarity 0.86)
- LLM-SRBench: A New Benchmark for Scientific Equation Discovery with Large Language Models (ICML 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
