AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

ProofBridge: Auto-Formalization of Natural Language Proofs in Lean via Joint Embeddings

Prithwish Jana, Kaan Kale, Ahmet Ege Tanriverdi, Cruise Song, Sriram Vishwanath, Vijay Ganesh

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Text generation and evaluation metrics · 16 citations (OpenAlex)

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 changedWhereMeasuredEffect
add retrieved relevant FL examples to SFT input (retrieval-augmented SFT) (ablation)ProofBridge (Kimina-Prover-RL-1.7B) · miniF2F-Test-PFSemantic correctness pass@32 (%): 34.84 -> 55.33Improved
add retrieved relevant FL examples to SFT input (retrieval-augmented SFT) (ablation)ProofBridge (Kimina-Prover-RL-1.7B) · miniF2F-Test-PFType correctness pass@32 (%): 78.69 -> 89.75Improved
add iterative proof repair with Lean feedback and equivalence feedback (ablation)ProofBridge (Retrieval-augmented SFT) · miniF2F-Test-PFSemantic correctness pass@32 (%): 55.33 -> 62.70Improved
add iterative proof repair with Lean feedback and equivalence feedback (ablation)ProofBridge (Retrieval-augmented SFT) · miniF2F-Test-PFType correctness pass@32 (%): 89.75 -> 95.49Improved
fine-tune with SFT on NL-FL pairs, evaluated with joint-embedding few-shot examples (ablation)Kimina-Prover-RL-1.7B (text-based retrieval few-shot) · miniF2F-Test-PFSC / TC (relative change): +2.06% SC, -11.06% TCMixed
add five random few-shot examplesKimina-Prover-RL-1.7B · miniF2F-Test-PFSemantic correctness pass@32 (%): 40.16 -> 31.56Hurt
add five random few-shot examplesKimina-Prover-RL-1.7B · miniF2F-Test-PFType correctness pass@32 (%): 75.00 -> 93.85Improved
add text-based retrieved few-shot examples (Qwen3-Embedding-8B) vs random few-shotKimina-Prover-RL-1.7B · miniF2F-Test-PFSC / TC pass@32 (%): SC 31.56 -> 32.78; TC 93.85 -> 89.75Mixed
replace baseline NL encoder with contrastively trained joint NL-FL embeddingNL->FL retrieval · NuminaMath-Lean-PF testRecall@1 (%): 16.06 -> 52.83Improved
use joint embedding instead of larger SoTA encoder Qwen3-Embedding-8BNL->FL retrieval · NuminaMath-Lean-PF testRecall@1 (%): 46.75 -> 52.83Improved
two-step pipeline (theorem autoformalizer then prover) vs end-to-endHerald-Translator -> Kimina-Prover-Distill-8B · miniF2F-Test-PFType correctness pass@32 (%): 59.43Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy