Core Concepts
Deep learning, especially the emergence of large language models, has sparked a notable surge of research exploring techniques to enhance the process of automated theorem proving.
Abstract
This survey provides a comprehensive overview of the current research landscape in deep learning for theorem proving. It covers the following key aspects:
Background on informal and formal theorem proving, including the concepts of automated theorem proving (ATP) and interactive theorem proving (ITP).
Detailed discussion of the various tasks and methods within this domain, such as autoformalization, premise selection, proofstep generation, proof search, and other related tasks.
Review of the available datasets for theorem proving, including both manually curated and synthetically generated datasets.
Evaluation of the metrics used in existing methods and an assessment of their performance.
Identification of the prevailing challenges in this field, including data scarcity, disunified evaluation protocols, and the need for better human-AI interaction.
Exploration of promising future directions, such as conjecturing, verified code generation, and the integration of theorem proving into math education.
The survey aims to serve as a foundational reference for deep learning approaches in theorem proving, seeking to catalyze further research endeavors in this rapidly growing field.
Stats
"Theorem proving is a cornerstone of mathematics."
"The recent development of deep learning, especially with the evolution of large language models (LLMs), has ignited a wave of research interest in this area again."
"The volume of papers on deep learning for theorem proving has grown approximately from 2 in 2016 to 50 in 2023."
Quotes
"Proving theorems is a cornerstone of mathematics."
"The recent development of deep learning, especially with the evolution of large language models (LLMs), has ignited a wave of research interest in this area again."
"Our survey aims to serve as a foundational reference for deep learning approaches in theorem proving, seeking to catalyze further research endeavors in this rapidly growing field."