ImProver 2: Iteratively Self-Improving LMs for Neurosymbolic Proof Optimization

Riyaz Ahuja, Tate Rowney, Jeremy Avigad, Sean Welleck

arXiv:2605.22885 · 2026-05-25 공개 · arXiv · PDF

data-efficient neurosymbolic lean-4 formal-mathematics proof-optimization model-scaffolding training-metrics expert-iteration

Abstract

Formal mathematics libraries are rapidly expanding, creating a growing need to refactor verified proofs for maintainability and to improve training data quality for neural provers. However, scalable proof optimization is hindered by heterogeneous and heuristically specified objectives, scarce data, and high training and inference costs. To overcome these challenges, we introduce ImProver 2, a neurosymbolic framework for automated proof optimization in Lean 4. ImProver 2 combines a data-efficient expert-iteration pipeline with a scaffold that exposes formal structure alongside lightweight informal abstractions. We further introduce a suite of metrics capturing structural proof properties. Using ImProver 2, we train a 7B-parameter model that outperforms orders-of-magnitude larger models within the same model family, and is competitive with mid-tier frontier models across metrics. We additionally demonstrate that our neurosymbolic scaffold significantly improves performance across both small and frontier models. We show that with proper scaffolding and training, small models can effectively restructure research-level proofs over complex and varied metrics, matching substantially larger systems and establishing proof optimization as a scalable, learnable task.

한국어 요약

📋 한 줄 요약

**[Lean 4 Proof Optimization]** ImProver 2가 expert-iteration·neurosymbolic scaffold·구조 메트릭으로 7B 모델을 학습 — 동일 family의 훨씬 큰 모델 outperform·중급 frontier 모델 수준의 자동 증명 최적화.

🎯 핵심 기여도

💡 핵심 아이디어

Proof optimization은 단순 텍스트 생성으로 환원할 수 없으며 formal 구조와 informal 추상을 함께 노출하는 scaffold + expert iteration의 결합이 작은 모델로도 research-level 증명을 재구조화할 수 있게 한다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: Proof optimization을 scalable·learnable task로 정립, 작은 모델로 대형 모델 성능 달성하는 cost-efficiency, formal+informal scaffold의 일반화 가능 설계 패턴 제시. **한계**: Lean 4 특화 — 다른 정형 시스템(Coq·Isabelle) 일반화 필요, 메트릭 suite의 평가 일관성, expert-iteration의 데이터 효율성 한계.

🚀 실용적 활용