Holophrasm: a neural Automated Theorem Prover for higher-order logic

TL;DR

Holophrasm is a neural-based automated theorem prover for higher-order logic that leverages deep learning, the Metamath formalism, and a neural-augmented bandit algorithm to prove theorems without hand-crafted features.

Contribution

It introduces a novel neural theorem proving system that combines deep learning with a bandit algorithm for exploring proof trees in higher-order logic.

Findings

Proves 14% of test theorems from Metamath's set.mm module.

Uses a sequence-to-sequence model for action enumeration.

Employs a neural-network-augmented bandit algorithm for proof search.

Abstract

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.