Summary of Learning Formal Mathematics From Intrinsic Motivation, by Gabriel Poesia et al.
Learning Formal Mathematics From Intrinsic Motivation
by Gabriel Poesia, David Broman, Nick Haber, Noah D. Goodman
First submitted to arxiv on: 30 Jun 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Logic in Computer Science (cs.LO)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper explores the Platonic view of mathematics as a discovery process, where mathematicians conjecture and prove theorems. The authors introduce Minimo, an agent that jointly learns to pose mathematical problems (conjecturing) and solve them (theorem proving). Given a mathematical domain axiomatized in dependent type theory, Minimo combines constrained decoding and type-directed synthesis to sample valid conjectures from a language model. The agent uses the same model to guide proof search, targeting hard but provable conjectures. Novel methods for hindsight relabeling on proof search trees improve the agent’s efficiency in both tasks. Experiments demonstrate that Minimo can bootstrap from axioms, self-improving in generating true and challenging conjectures and finding proofs. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about how people discovered math. The authors created a special computer program called Minimo that learns to come up with new math problems and solve them. They used a special kind of math language to make sure the program only came up with correct answers. The program got better at coming up with hard but solvable math problems as it learned. The authors tested Minimo on three different areas of math and found that it could learn to come up with new problems and solutions all on its own, starting from scratch. |
Keywords
» Artificial intelligence » Language model
