Skip to yearly menu bar Skip to main content


Virtual presentation / top 25% paper

Formal Mathematics Statement Curriculum Learning

Stanislas Polu · Jesse Han · Kunhao Zheng · Mantas Baksys · Igor Babuschkin · Ilya Sutskever

Keywords: [ expert iteration ] [ language modeling ] [ neural theorem proving ] [ formal mathematics ] [ Machine Learning for Sciences ]


Abstract:

We explore the use of expert iteration in the context of language modeling applied to formal mathematics. We show that at same compute budget, expert iteration, by which we mean proof search interleaved with learning, dramatically outperforms proof search only. We also observe that when applied to a collection of formal statements of sufficiently varied difficulty, expert iteration is capable of finding and solving a curriculum of increasingly difficult problems, without the need for associated ground-truth proofs. Finally, by applying this expert iteration to a manually curated set of problem statements, we surpass previous state-of-the-art on the miniF2F benchmark, automatically solving multiple challenging problems drawn from high school olympiads.

Chat is not available.