Reproduce “Magnushammer: A Transformer-Based Approach to Premise Selection” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Maciej Mikuła, Szymon Tworkowski, Szymon Antoniak, Bartosz Piotrowski, Albert Qiaochu Jiang, Jin Peng Zhou, Christian Szegedy, Łukasz Kuciński and 2 more
Magnushammer, a transformer-based system, advances automated mathematical reasoning by enhancing premise selection in interactive theorem provers. It achieves a 59.5% proof rate on the PISA benchmark, surpassing Sledgehammer's 38.3%, and increases the success rate of neural-symbolic provers to 71.0%.
Key insights
- Transformer architectures, fine-tuned with contrastive learning on task-specific datasets, can learn robust representations for premise relevance, reducing reliance on extensive domain-specific engineering.
- A two-stage retrieval and re-ranking mechanism effectively balances computational efficiency and accuracy for premise selection in formal mathematics.
- Integrating neural premise selection into neural-symbolic theorem provers significantly boosts overall proof automation capabilities, demonstrating a synergistic effect.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.