Research map / Large language models / Agents, code and math
DecAEvolve: Decompose, Adapt, and Evolve for Effective LLM-based Scientific Equation Discovery
Pouya Behzadifar, Parshin Shojaee, Sanchit Kabra, Kazem Meidani, Chandan K. Reddy
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 GRPO test-time adaptation to LLM-SR (ablation) | LLM-SR · Oscillator, E. coli, Stress-Strain benchmarks, six backbones | best NMSE vs search candidates | Improved |
| add term-level decomposition feedback to LLM-SR (ablation) | LLM-SR · Oscillator, E. coli, Stress-Strain benchmarks, six backbones | best NMSE vs search candidates | Improved |
| combine decomposition and GRPO adaptation (ablation) | LLM-SR · benchmark problems, all backbones | terminal NMSE | Improved |
| use DecAEvolve instead of LLM-SR | Qwen2.5-7B · Oscillation 1 OOD | NMSE: 0.0017 -> 1.51e-5 | Improved |
| use DecAEvolve instead of LLM-SR | Qwen2.5-7B · E. coli growth OOD | NMSE: 0.1285 -> 0.0012 | Improved |
| use DecAEvolve instead of LLM-SR | Llama-3.1-8B · Stress-Strain ID | NMSE: 0.0376 -> 0.0144 | Improved |
| use DecAEvolve instead of LLM-SR | Qwen2.5-1.5B · E. coli growth ID | NMSE: 0.7237 -> 0.6719 | No clear effect |
| use DecAEvolve instead of LLM-SR | Qwen2.5-3B · E. coli growth OOD | NMSE: 0.8011 -> 0.0454 | Improved |
| use small DecAEvolve backbones instead of larger LLM-SR backbones | LLM-SR with Mixtral / GPT-3.5 · benchmarks | NMSE | Mixed |
| scale backbone from small to larger models | DecAEvolve · benchmarks | NMSE | Improved |
Closest papers by meaning
- LLM-SR: Scientific Equation Discovery via Programming with Large Language Models (ICLR 2025 · similarity 0.91)
- LLM-SRBench: A New Benchmark for Scientific Equation Discovery with Large Language Models (ICML 2025 · similarity 0.90)
- FunctionEvolve: Structure-Guided Symbolic Regression with LLMs (NeurIPS 2026 · similarity 0.88)
- Evolution Fine-Tuning: Learning to Discover Across 371 Optimization Tasks (NeurIPS 2026 · similarity 0.88)
- MOOSE-Chem2: Exploring LLM Limits in Fine-Grained Scientific Hypothesis Discovery via Hierarchical Search (NeurIPS 2025 · similarity 0.87)
- Symbolic Regression with a Learned Concept Library (NeurIPS 2024 · 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.
