Long-term reward value coding by midbrain dopamine neurons

我们发现，中脑多巴胺神经元编码长期价值及其对未来奖励的预期误差。采用电生理学方法和基于强化学习理论的计算机模拟方法，记录并分析了猴子在完成一个通过多步动作获得多个奖励的行为任务时多巴胺神经元的反应。我们还证实，多巴胺神经元通过学习任务来进化这些活动。这些功能在决策和行动选择中发挥着重要作用，以实现一个遥远的目标。

We showed that midbrain dopamine neurons encode long-term value and its expectation errors of future rewards. By using electrophysiological methods and computational simulations based on reinforcement learning theories, we recorded and analyzed dopamine neuron responses in monkeys performing a behavioral task in which they obtained multiple rewards by multi-step actions. We also confirmed that dopamine neurons evolved these activities by learning the task. These functions play an important role in decision-making and action selection to achieve a distant goal.