Holophrasm: a neural Automated Theorem Prover for higher-order logic
About
I propose a system for Automated Theorem Proving in higher order logic using deep learning and eschewing hand-constructed features. Holophrasm exploits the formalism of the Metamath language and explores partial proof trees using a neural-network-augmented bandit algorithm and a sequence-to-sequence model for action enumeration. The system proves 14% of its test theorems from Metamath's set.mm module.
Daniel Whalen• 2016
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Theorem Proving | set.mm (test) | Proofs Found (Test)557 | 14 | |
| Theorem Proving (Relevance Ranking) | set.mm (val) | Top-1 Accuracy51.52 | 12 | |
| Theorem Proving (Substitution Network) | set.mm (val) | Prob0.6142 | 12 | |
| Theorem Proving | iset.mm (test) | Proofs Found378 | 2 | |
| Theorem Proving (Relevance Ranking) | iset.mm (val) | Top-1 Accuracy43.27 | 2 | |
| Theorem Proving (Substitution Network) | iset.mm (val) | Probability0.1723 | 2 |
Showing 6 of 6 rows