Research map / Large language models / Agents, code and math
Towards Diverse Scientific Hypothesis Search with Large Language Models
Haorui Wang, Parshin Shojaee, Kazem Meidani, Kunyang Sun, José Miguel Hernández-Lobato, Teresa Head-Gordon, Jiajun He, Chandan K. Reddy, Chao Zhang, Yuanqi Du
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 hot pool with Metropolis-Hastings swaps | MOLLEO (DeepSeek-V3.2) · JNK3 | Top-10 AUC: 0.58 -> 0.63 | Improved |
| add hot pool with swaps | MOLLEO · JNK3 | Top-10 Avg: 0.66 -> 0.74 | Improved |
| add hot pool with swaps | MOLLEO · GSK3β | Top-10 AUC: 0.70 -> 0.76 | Improved |
| add hot pool with swaps | MOLLEO · GSK3β | Top-10 Avg: 0.82 -> 0.82 | No clear effect |
| add hot pool with swaps | MOLLEO · JNK3 | number of diverse candidates: ~50 -> ~90 | Improved |
| use single-population high-temperature LLM sampling (Tempering) instead of base MOLLEO (ablation) | MOLLEO · JNK3 | Top-10 AUC: 0.58 -> 0.46 | Hurt |
| use two isolated pools (Ensemble) rather than swapping (ablation) | EvoDiverse · JNK3 | Top-10 AUC: 0.63 -> 0.54 | Hurt |
| add parallel tempering to GraphGA | GraphGA · GSK3β | Top-10 AUC: 0.59 -> 0.67 | Improved |
| add parallel tempering to GraphGA | GraphGA · JNK3 | Top-10 Avg: 0.59 -> 0.52 | Hurt |
| add hot pool with swaps | Ensemble (LLM-SR) · LSR-SRBench Physics | Best Acc0.1: 0.283 -> 0.408 | Improved |
| add hot pool with swaps | Ensemble (LLM-SR) · LSR-SRBench Biology | Best Acc0.1: 0.104 -> 0.212 | Improved |
| add hot pool with swaps | Ensemble (LLM-SR) · LSR-SRBench Materials | Top-10 Acc0.1: 0.757 -> 0.763 | No clear effect |
Closest papers by meaning
- MOOSE-Chem2: Exploring LLM Limits in Fine-Grained Scientific Hypothesis Discovery via Hierarchical Search (NeurIPS 2025 · similarity 0.91)
- Efficient Evolutionary Search Over Chemical Space with Large Language Models (ICLR 2025 · similarity 0.88)
- Evolution Fine-Tuning: Learning to Discover Across 371 Optimization Tasks (NeurIPS 2026 · similarity 0.87)
- LLM-SR: Scientific Equation Discovery via Programming with Large Language Models (ICLR 2025 · similarity 0.87)
- Measuring Scientific Capabilities of Language Models with a Systems Biology Dry Lab (NeurIPS 2025 · similarity 0.86)
- FunctionEvolve: Structure-Guided Symbolic Regression with LLMs (NeurIPS 2026 · similarity 0.85)
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.
