Research map / Large language models / Instruction and fine-tuning
Divide and Abstract: Autoformalization via Decomposition and Abstraction Learning
Marcus J. Min, Yeqi Gao, Wilson Sy, Zhaoyu Li, Xujie Si, Osbert Bastani
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 learned abstractions and decomposition (full DNA) to the baseline | non-reasoning models average · LeanEuclidPlus | pass@1 accuracy: 18.7 -> 40.6 | Improved |
| add learned abstractions and decomposition (full DNA) to the baseline | reasoning models average · LeanEuclidPlus | pass@1 accuracy: 31.5 -> 51.1 | Improved |
| add decomposition (Divide) to the baseline (ablation) | non-reasoning models · LeanEuclidPlus | pass@1 accuracy: 18.7 -> 28.9 | Improved |
| add learned abstractions (Abstract) to the baseline (ablation) | non-reasoning models · LeanEuclidPlus | pass@1 accuracy: 18.7 -> 20.9 | Improved |
| add decomposition on top of learned abstractions (Abstract -> DNA) (ablation) | non-reasoning models · LeanEuclidPlus | pass@1 accuracy: 20.9 -> 40.6 | Improved |
| add learned abstractions on top of decomposition (Divide -> DNA) (ablation) | reasoning models · LeanEuclidPlus | pass@1 accuracy: 36.5 -> 51.1 | Improved |
| add decomposition to Qwen3-14B | Qwen3 14B · LeanEuclidPlus | pass@1 accuracy: 1.0 -> 9.6 | Improved |
| add DNA to baseline on ProofNet-Hard | GPT-5 mini · ProofNet-Hard | pass@1 accuracy: 0.0 -> 22.1 (DNOracleA) | Improved |
| add decomposition prompts to specialized autoformalizers (Abstract -> DNA) (ablation) | Kimina Autoformalizer 7B · ProofNet-Hard | pass@1 accuracy: 8.4 -> 1.1 | Hurt |
| add decomposition to abstractions on harder problems (Abstract -> DNA) (ablation) | gpt-5-mini · LeanEuclidPlus (40 complex) | pass@1 accuracy: 47.5 -> 57.5 | Improved |
| add 1-shot example (vs 0-shot) to DNA (ablation) | non-reasoning models average · LeanEuclidPlus | pass@1 accuracy: 19.3 -> 40.6 | Improved |
| add 1-shot example (vs 0-shot) to DNA (ablation) | reasoning models average · LeanEuclidPlus | pass@1 accuracy: 26.0 -> 51.1 | Improved |
Closest papers by meaning
- Autoformalize Mathematical Statements by Symbolic Equivalence and Semantic Consistency (NeurIPS 2024 · similarity 0.89)
- ATLAS: Autoformalizing Theorems through Lifting, Augmentation, and Synthesis of Data (NeurIPS 2025 · similarity 0.89)
- Multi-language Diversity Benefits Autoformalization (NeurIPS 2024 · similarity 0.88)
- FormalAlign: Automated Alignment Evaluation for Autoformalization (ICLR 2025 · similarity 0.88)
- APOLLO: Automated LLM and Lean Collaboration for Advanced Formal Reasoning (NeurIPS 2025 · similarity 0.87)
- ImProver: Agent-Based Automated Proof Optimization (ICLR 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.
