Towards Finding Longer Proofs
This work addresses the problem of scaling automated theorem proving to longer proofs, which is incremental as it builds on existing RL methods but focuses on data efficiency and confidence.
The paper tackles the challenge of automated theorem proving by developing a reinforcement learning guidance system, FLoP, which generalizes from minimal training data to find long proofs, achieving competitive performance on benchmarks with very long proofs.
We present a reinforcement learning (RL) based guidance system for automated theorem proving geared towards Finding Longer Proofs (FLoP). Unlike most learning based approaches, we focus on generalising from very little training data and achieving near complete confidence. We use several simple, structured datasets with very long proofs to show that FLoP can successfully generalise a single training proof to a large class of related problems. On these benchmarks, FLoP is competitive with strong theorem provers despite using very limited search, due to its ability to solve problems that are prohibitively long for other systems.