Multi-agent Social Reinforcement Learning Improves Generalization - 42Papers