Imbalanced Datasets

The paper discusses the challenges of offline reinforcement learning (RL) when dealing with imbalanced datasets that are dominated by suboptimal trajectories. It finds that current offline RL algorithms tend to…