This repository contains files used for final data preprocessing, fine-tuning and evaluation of BERT models (and a single Wavenet) used to solve a theorem proving task. Code is integrated with Deephol theorem prover and evaluation is run in a HOList benchmark. These codes were used during the preparation of the Bachelor's thesis "Using BERT and its modifications in the DeepHOL theorem prover".
The thesis is based on the following two papers:
- "HOList: An Environment for Machine Learning of Higher-Order Theorem Proving", https://arxiv.org/pdf/1904.03241.pdf
- "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding", https://arxiv.org/pdf/1810.04805.pdf
As a result this repository contains parts of code from the respective repositories:
- When we mention Deephol/Holist we refer to: https://github.com/tensorflow/deepmath
- When we mention Bert we refer to: https://github.com/google-research/bert