DeepReinforcementLearningattheEdgeofthe StatisticalPrecipice

Research in artificial intelligence, and particularly deep reinforcement learning (RL), relies on evaluating aggregate performance on a diverse suite of tasks to assess progress.