EfficientSchedulingofDataAugmentation forDeepReinforcementLearning

However,evenwhentheprior is useful for generalization, distilling it to RL agent often interferes with RL training and degenerates sample efficiency.