The general trend in machine learning research is to stop fine-tuning models, and instead use a meta-learning algorithm that automatically finds the best architecture and hyperparameters. What about meta-reinforcement learning (meta-RL)? Meta-RL is just meta-learning applied to RL.
