Research map / Large language models / Agents, code and math
DRIFT: Decompose, Retrieve, Illustrate, then Formalize Theorems
Meiru Zhang, Philipp Borchert, Milan Gritta, Gerasimos Lampouras
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 query decomposition before retrieval (ablation) | dense retriever · ProofNet dependency retrieval | F1: 13.77 -> 27.68 | Improved |
| add query decomposition before retrieval (ablation) | dense retriever · MiniF2F-test dependency retrieval | F1: 0.66 -> 3.83 | Improved |
| add query decomposition before retrieval (ablation) | dense retriever · ConNF dependency retrieval | F1: 28.17 -> 36.88 | Improved |
| add Illustrate module (illustrative theorems) (ablation) | DRIFT GPT-4.1 · ConNF | BEq+@1: 35.90 -> 54.84 | Improved |
| add Illustrate module (illustrative theorems) (ablation) | DRIFT GPT-4.1 · ProofNet | BEq+@1: 14.17 -> 17.38 | Improved |
| add Decompose module on top of Illustrate (ablation) | DRIFT GPT-4.1 · ProofNet | BEq+@1: 13.64 -> 17.38 | Improved |
| add Decompose module (ablation) | DRIFT GPT-4.1 · MiniF2F-test | BEq+@1: 26.34 -> 24.55 | Hurt |
| add retrieval of premises (ablation) | DRIFT GPT-4.1 · ConNF | BEq+@1: 4.47 -> 54.84 | Improved |
| use DRIFT instead of DPR (RAuto) | GPT-4.1 · ProofNet | BEq+@10: 19.25 -> 21.93 | Improved |
| use DRIFT instead of zero-shot | GPT-4.1 · ConNF | BEq+@10: 6.76 -> 62.33 | Improved |
| use DRIFT instead of DPR (RAuto) | GPT-4.1 · MiniF2F-test | BEq+@10: 32.14 -> 29.02 | Hurt |
| add DRIFT retrieved context | DeepSeek-V3.1 · MiniF2F-test | BEq+@10: 27.23 -> 24.55 | Hurt |
Closest papers by meaning
- ATLAS: Autoformalizing Theorems through Lifting, Augmentation, and Synthesis of Data (NeurIPS 2025 · similarity 0.90)
- Autoformalize Mathematical Statements by Symbolic Equivalence and Semantic Consistency (NeurIPS 2024 · similarity 0.90)
- RealMath: A Continuous Benchmark for Evaluating Language Models on Research-Level Mathematics (NeurIPS 2025 · similarity 0.90)
- Putnam-AXIOM: A Functional & Static Benchmark for Measuring Higher Level Mathematical Reasoning in LLMs (ICML 2025 · similarity 0.89)
- FaithSieve: Fine-Grained Evaluation of Math Proofs with Faithful Formal Evidence (NeurIPS 2026 · similarity 0.89)
- Semantic Search over 9 Million Mathematical Theorems (NeurIPS 2026 · similarity 0.89)
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.
