Verifiably Safe Off-Model Reinforcement Learning

Nathan Fulton, Andre Platzer

The desire to use reinforcement learning in safety-critical settings has inspired a recent interest in formal methods for learning algorithms. Existing formal methods for learning…