Research map / Large language models / Agents, code and math
Unleashing Scientific Reasoning for Bio-experimental Protocol Generation via Structured Component-based Reward Mechanism
Haoran Sun, Yankai Jiang, Zhenyu Tang, Yaning Pan, Shuang Gu, Zekai Lin, Lilong Wang, Wenjie Lou, Lei Liu, LEI BAI, Xiaosong Wang
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 step-count reward f(d) (ablation) | Thoth · SciRecipe-Eval | Step-M: 10.00 -> 53.00 | Improved |
| add length penalty g(L) (ablation) | Thoth · SciRecipe-Eval | Semantic-A: 38.89 -> 46.60 | Improved |
| add positional decay factor m_ij (ablation) | Thoth · SciRecipe-Eval | Semantic-A: 38.68 -> 46.60 | Improved |
| add Order consistency reward (ablation) | Thoth · SciRecipe-Eval | Order-S: 12.83 -> 25.50 | Improved |
| add KL penalty (ablation) | Thoth · SciRecipe-Eval | AVG: 48.89 -> 52.10 | Improved |
| SCORE reward instead of vanilla lexical reward (ablation) | Thoth · SciRecipe-Eval | AVG: 45.62 -> 52.10 | Improved |
| SCORE reward instead of vanilla lexical reward (ablation) | Thoth · SciRecipe-Eval | BLEU-AVG: 45.52 -> 43.62 | Hurt |
| add scientific review (SciCheck) data (ablation) | Thoth · SciRecipe-Eval | AVG: 47.37 -> 49.45 | Improved |
| add Sketch-and-Fill paradigm (ablation) | Qwen3-8B · SciRecipe-Eval | BLEU-AVG | No clear effect |
| add Sketch-and-Fill paradigm (ablation) | DeepSeek-V3, GPT-5 Chat, Thoth · SciRecipe-Eval | lexical similarity avg: +3.89%, +2.92%, +3.79% | Improved |
| three-stage training (pretrain+SFT+RL) vs Stage 1+2 (ablation) | Thoth · SciRecipe-Eval | executability avg: +11.08% | Improved |
Closest papers by meaning
- MOOSE-Chem2: Exploring LLM Limits in Fine-Grained Scientific Hypothesis Discovery via Hierarchical Search (NeurIPS 2025 · similarity 0.87)
- Measuring Scientific Capabilities of Language Models with a Systems Biology Dry Lab (NeurIPS 2025 · similarity 0.85)
- SciCode: A Research Coding Benchmark Curated by Scientists (NeurIPS 2024 · similarity 0.85)
- SciInstruct: a Self-Reflective Instruction Annotated Dataset for Training Scientific Language Models (NeurIPS 2024 · similarity 0.85)
- Protein Design with Agent Rosetta: A Case Study for Specialized Scientific Agents (ICML 2026 · similarity 0.85)
- CURIE: Evaluating LLMs on Multitask Scientific Long-Context Understanding and Reasoning (ICLR 2025 · 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.
