Research map / Large language models / Agents, code and math
Towards Multimodal Data-Driven Scientific Discovery Powered by LLM Agents
Fan Liu, Xiaozhao Zeng, Hao Liu
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 task-provided domain knowledge (ablation) | ReAct (o4-mini) · MoSciBench avg | accuracy: 48.4% -> 44.9% | Hurt |
| add workflow scaffolding (ablation) | ReAct (o4-mini) · MoSciBench avg | accuracy: 48.4% -> 54.1% | Improved |
| add workflow scaffolding (ablation) | ReAct (o4-mini) · MoSciBench Climate Sci. | accuracy: 57.1% -> 71.4% | Improved |
| add workflow scaffolding (ablation) | ReAct (o4-mini) · MoSciBench Earth Sci. | accuracy: 35.7% -> 50.0% | Improved |
| add domain knowledge (ablation) | ReAct (o4-mini) · MoSciBench Climate Sci. | accuracy: 57.1% -> 50.0% | Hurt |
| add domain knowledge (ablation) | ReAct (o4-mini) · MoSciBench Cheminformatics | accuracy: 33.3% -> 26.7% | Hurt |
| add workflow scaffolding (ablation) | ReAct (o4-mini) · MoSciBench alignment errors | accuracy: 31.8% -> 27.3% | Improved |
| give agent data access instead of NoDataGuess | o4-mini · MoSciBench overall | accuracy: 0.105 -> 0.484 | Improved |
| use stronger base model (gpt-5-mini to o4-mini) | ReAct · MoSciBench overall | accuracy: 0.174 -> 0.484 | Improved |
| add retrieval of external knowledge (RAG-ReAct vs ReAct) | gpt-5-mini · MoSciBench overall | accuracy: 0.174 -> 0.128 | Hurt |
| add oracle-feedback retries (Reflexion vs ReAct) | o4-mini · MoSciBench overall | accuracy: 0.484 -> 0.458 | Hurt |
| increase Best-of-N rollouts beyond N = 3 | ReAct (DeepSeek-V3.1) · MoSciBench | accuracy | Mixed |
Closest papers by meaning
- ScienceBoard: Evaluating Multimodal Autonomous Agents in Realistic Scientific Workflows (ICLR 2026 · similarity 0.90)
- PhysGym: Benchmarking LLMs in Interactive Physics Discovery with Controlled Priors (NeurIPS 2025 · similarity 0.89)
- DiscoveryBench: Towards Data-Driven Discovery with Large Language Models (ICLR 2025 · similarity 0.88)
- CURIE: Evaluating LLMs on Multitask Scientific Long-Context Understanding and Reasoning (ICLR 2025 · similarity 0.88)
- MOOSE-Chem2: Exploring LLM Limits in Fine-Grained Scientific Hypothesis Discovery via Hierarchical Search (NeurIPS 2025 · similarity 0.88)
- AstaBench: Rigorous Benchmarking of AI Agents with a Scientific Research Suite (ICLR 2026 · similarity 0.88)
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.
