Experiments towards neural network theorem proving
The Deepmath project seeks to improve automated theorem proving using deep learning and other machine learning techniques. Deepmath is a collaboration between Google Research and several universities.
The source code in this repository is not an official Google product, but is a research collaboration with external research teams.
Deepmath depends on TensorFlow, which is included as a submodule. Use, or
see, the Dockerfile for build instructions for deephol, our neural prover. It
requires connecting to a proof assistant server. See
https://github.com/brain-research/hol-light for a server implementation.
No open issues yet, or sync has not completed.