The influence of top-down control on feedback processing and reinforcement learning in decision tasks

对外部反馈的评估是做出最优决策的重要前提。反馈可以通过考虑过去决策的后果来改进未来的决策。近年来，反馈学习被纳入强化学习的理论框架。也就是说，反馈一直被视为导致决策中涉及的刺激-反应关联的调整的增强剂。这一点的证据来自使用事件相关电位(ERPs)检查反馈的研究。反馈相关负性(FRN)是由负反馈引起的一种负偏向，它反映了一种预测误差，该误差被用作行为调整的学习信号。本项目旨在调查自上而下控制是否以及如何影响FRN--从而由于反馈而导致的强化学习。在计划中的实验中，参与者被告知反馈的有效性。这创造了这样一种情况，即从反馈中学习或多或少是有益的，因此，强化学习应该或多或少受到自上而下控制的调节。通过将强化学习的形式模型应用于经验数据，我们的目标是描述这些自上而下效应与强化学习之间的相互作用。将进行进一步的实验，以揭示主动控制或选择性注意等机制如何有助于自上而下的控制。一方面，我们的研究旨在回答强化学习是否会受到认知过程的影响这一经典问题。另一方面，我们希望通过考虑认知过程来验证和改进现有的强化学习模型。

Evaluation of external feedback is an important prerequisite for optimal decision-making. Feedback allows for improving future decisions by taking the consequences of past decisions into account. In recent years, learning from feedback has been considered within the theoretical framework of reinforcement learning. That is, feedback has been viewed as a reinforcer that leads to the adjustment of stimulus-response associations involved in a decision. Evidence for this comes from studies that examined feedback using event-related potentials (ERPs). The feedback-related negativity (FRN), a negative deflection elicited by negative feedback, has been shown to reflect a prediction error that is used as a learning signal for the adjustment of behavior. The present project aims to investigate whether and how the FRN - and thus reinforcement learning due to feedback - is influenced by top-down control. In the planned experiments, participants are instructed about the validity of feedback. This creates a situation in which learning from feedback is more or less beneficial, and hence, in which reinforcement learning should be more or less modulated by top-down control. By applying formal models of reinforcement learning to empirical data, we aim to describe the interaction between these top-down effects and reinforcement learning. Further experiments will be conducted to reveal how mechanisms like proactive control or selective attention contribute to top-down control. On the one hand, our studies aim to provide an answer to the classical question whether reinforcement learning can be influenced by cognitive processes. On the other hand, we hope to validate and improve current models of reinforcement learning by taking cognitive processes into account.