Single-trial reward learning and generalization

描述(由申请者提供)：为了适应行为，个人必须学习奖励结果与环境特征之间的联系，并利用这些联系来指导在新情况下的行为。先前在动物模型中的研究表明，在联想学习过程中参与的神经系统可以影响这种概括行为，因此适应性概括需要海马体的参与。然而，大多数关于人类受试者的研究都研究了纹状体引导的奖赏学习和泛化，通常是在多次尝试的概率反馈学习过程中。在日常生活中，奖赏价值往往是在一次经历或经历之后获得的；指导这种单次尝试奖赏反馈学习的神经机制尚不清楚。本研究提出单次奖赏反馈学习和泛化的特征可能是强化学习和情景记忆系统之间的相互作用。我们将通过结合行为方法、计算建模和高分辨率神经成像来验证这一说法。我们将使用高分辨率功能磁共振成像(目标1)来表征在获取单次试验奖励反馈关联期间参与的神经系统。此外，我们将使用强化学习建模(Aim 2)和基于模型的功能磁共振成像(Aim 3)来表征通过单次尝试奖励反馈的联想学习如何有助于泛化行为。我们预测(1)单次尝试奖赏反馈学习将在联想学习中参与纹状体和海马体之间的相互作用，以及(2)这两个系统之间的相互作用将促进奖赏联系的适应性概括。从这一建议中获得的知识将促进我们对奖赏学习的神经生物学的理解，并为健康和受损的奖赏相关行为(如吸毒成瘾)的机制提供证据。

DESCRIPTION (provided by applicant): To behave adaptively, individuals must learn associations between reward outcomes and features of the environment, and use these associations to inform behavior in novel situations. Previous research in animal models has demonstrated that the neural systems engaged during associative learning can influence such generalization behaviors, such that adaptive generalization requires engagement of the hippocampus. However, the majority of research in human subjects has investigated striatal-guided reward learning and generalization, often during multi-trial, probabilistic feedback learning. In everyday life, reward values are often acquired after only a single experience or episode; the neural mechanisms guiding such single-trial reward feedback learning are undefined. The current proposal advances the claim that single-trial reward feedback learning and generalization may be characterized by the interactions between reinforcement learning and episodic memory systems. We will test this claim by combining behavioral methods, computational modeling, and high-resolution neuroimaging. We will characterize the neural systems engaged during the acquisition of single-trial reward feedback associations using hi-resolution fMRI (Aim 1). Further, we will characterize how associative learning via single-trial reward feedback contributes to generalization behavior using reinforcement-learning modeling (Aim 2) and model-based fMRI (Aim 3). We predict that (1) single-trial reward-feedback learning will engage interactions between the striatum and hippocampus during associative learning, and (2) the interaction between these systems will promote adaptive generalization of reward associations. The knowledge obtained from this proposal will advance our understanding of the neurobiology of reward learning and provide evidence for the mechanisms underlying healthy and impaired reward-related behaviors such as drug addiction.