Enabling user-driven Checkpointing strategies in Reverse-mode Automatic Differentiation