Project Grant 2403211

This $611,492 Project Grant award from the National Science Foundation's (NSF) Computer and Information Science and Engineering (CISE) program to the University of Texas at Austin focuses on developing neurosymbolic AI techniques to enhance the accessibility and efficiency of interactive formal theorem provers. The goal is to automate the low-level aspects of theorem-proving, enabling wider use of formal verification tools for applications like safer software, more robust hardware, and improved mathematical rigor. Key research tasks include developing methods for training large language models on proof data, combining reinforcement learning and search for efficient inference, and automatically discovering proof tactics through proof compression. This work aims to create a powerful toolkit for automating many types of proofs that have traditionally been done manually, making interactive theorem provers significantly more usable. The project will also involve training graduate and undergraduate students at UT Austin in this emerging field that combines formal methods and machine learning.