Reproduce “Peano: Learning Formal Mathematical Reasoning” 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
Gabriel Poesia, Noah D. Goodman
This research from Stanford University introduces Peano, a novel theorem-proving environment, and demonstrates that equipping a reinforcement learning agent with the ability to induce reusable procedural abstractions (tactics) enables it to solve complex mathematical reasoning problems. The learned abstractions align with human-designed curricula and can accelerate the learning of subsequent agents, providing a computational account of mathematical knowledge transmission.
Key insights
- The ability to induce reusable procedural abstractions (tactics) is critical for overcoming combinatorial challenges in automated mathematical reasoning and making progress on complex problems.
- Learned computational abstractions can inherently structure a domain in a way that aligns with human pedagogical practices and expert-designed curricula.
- Structured curricula, derived from the dependencies of learned abstractions, can significantly accelerate the acquisition of mathematical knowledge for subsequent learners, computationally mirroring efficient cultural transmission.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.