Excluding the Irrelevant: Focusing Reinforcement Learning through Continuous Action Masking

Neural Information Processing Systems 

Focusing learning on these relevant actions can significantly improve training efficiency and effectiveness.