Research map / Large language models / Agents, code and math
LoC-Decomp: LLM Autoformalization via Logical Concept Decomposition and Iterative Feedback Correction
Jiangze Shi, Zhiwei Zhang, Baoquan Ma, Shuai Zhao, Ye Yuan, Guoren 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 iterative rectification (3 rounds) (ablation) | LoC-Decomp · MATH-500 (ASCC-3-MV) | pass rate: 63.20 -> 84.00 | Improved |
| add iterative rectification (3 rounds) (ablation) | LoC-Decomp · miniF2F (ASCC-3-MV) | pass rate: 77.25 -> 90.16 | Improved |
| add iterative rectification (3 rounds) (ablation) | LoC-Decomp · PutnamBench (ASCC-3-MV) | pass rate: 48.37 -> 81.50 | Improved |
| add iterative rectification (3 rounds) (ablation) | LoC-Decomp · MATH-Level5-50 (human eval) | pass rate: 70.00 -> 84.00 | Improved |
| use LoC-Decomp template and few-shot examples (no iteration) vs basic few-shot baseline (ablation) | DeepSeek-V3 · MATH-Level5-50 (human eval) | pass rate: 52.00 -> 70.00 | Improved |
| use ASCC semantic checker instead of basic semantic checker in the iterative loop (ablation) | iterative formalization · MATH-Level5-50 (human eval) | pass rate: 54.00 -> 84.00 | Improved |
| use ASCC-3-MV (majority vote) checker instead of plain SC-Baseline (ablation) | semantic consistency checking · MATH-ASCC-Eval-150 | precision: 0.68 -> 0.90 | Improved |
| compare LoC-Decomp with DeepSeek-Prover-V2 on ProofNet | LoC-Decomp · ProofNet (LLM-judge pass rate) | pass rate: 73.57 vs 83.10 | Hurt |
Closest papers by meaning
- Don't Trust: Verify -- Grounding LLM Quantitative Reasoning with Autoformalization (ICLR 2024 · similarity 0.90)
- Autoformalize Mathematical Statements by Symbolic Equivalence and Semantic Consistency (NeurIPS 2024 · similarity 0.90)
- APOLLO: Automated LLM and Lean Collaboration for Advanced Formal Reasoning (NeurIPS 2025 · similarity 0.90)
- Can LLMs Correct Themselves? A Benchmark of Self-Correction in LLMs (NeurIPS 2025 · similarity 0.88)
- CRANE: Reasoning with constrained LLM generation (ICML 2025 · similarity 0.88)
- Understanding and Mitigating Numerical Sources of Nondeterminism in LLM Inference (NeurIPS 2025 · 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.
