Interactive development of reinforcement learning and adaptive memory

项目摘要 焦虑和抑郁越来越被认为是起源于发展的疾病。而 易受焦虑和抑郁的影响在成年之前就增加了，这是导致焦虑和抑郁的发展因素。 对这种增加的风险还没有很好的理解。两个典型的学习和记忆偏差与 焦虑和抑郁的病因学：负价信息的优先加工和倾向 形成过于笼统的记忆和价值联想。尽管有明显的临床相关性， 将这些学习和记忆偏差与精神病风险联系起来的数据在很大程度上依赖于回忆测量 无法研究它们如何通过基于价值的学习和记忆在发展过程中产生 编码过程。在这个建议中，我们将利用计算建模和神经成像方法 阐明学习计算和记忆形成之间的机械关系是如何成为效价的基础的， 以及从童年到成年的过度概括偏见。目标1将描述如何 学习中的效价偏差随着发展而变化，它们如何影响对情节细节的偶然记忆 以及它们如何通过神经计算产生。目标2的特点是， 发展，学习表征的一般性如何适应上下文，记忆的特异性如何 表征随着时间的推移而变化，以及神经表征如何支持使用多个级别的 抽象来指导学习和记忆。目标3将描述效价和泛化偏差如何变化 并评估其与真实世界自传体记忆和临床 医学。该研究的意义在于：1）提供了一个理论上的 解释了基于价值的学习中的效价和泛化偏差以及情景学习中的相应偏差 和自传体记忆; 2）阐明这些偏见背后的神经认知机制; 3）描绘 从童年到成年，这些过程的规范性纵向发展变化; 4）建立 是否计算表型捕捉到这些偏见预测焦虑和抑郁症。

PROJECT SUMMARY Anxiety and depression are increasingly recognized as disorders that are developmental in origin. While vulnerability to anxiety and depression is heightened prior to adulthood, the developmental factors that give rise to this increased risk are not well understood. Two characteristic learning and memory biases are implicated in the etiology of anxiety and depression: preferential processing of negatively valenced information and a tendency to form overly general memories and value associations. Despite their apparent clinical relevance, studies to date linking these learning and memory biases to psychiatric risk have relied largely on recollective measures that do not enable the study of how they may arise over development through value-based learning and memory encoding processes. In this proposal, we will leverage computational modeling and neuroimaging approaches to elucidate how mechanistic relations between learning computations and memory formation underlie valence and overgeneralization biases across development from childhood to adulthood. Aim 1 will characterize how valence biases in learning change over development, how they influence incidental memory for episodic details of valenced outcomes, and how they arise through neural computations. Aim 2 will characterize, across development, how generality of learned representations adapts across contexts, how the specificity of memory representations changes with time, and how neural representations support the use of multiple levels of abstraction to guide learning and memory. Aim 3 will characterize how valence and generalization biases change longitudinally with age and assess their relation to real-world autobiographical memory and clinical symptomatology. The significance of the proposed research lies in its potential to: 1) provide a theoretical account relating valence and generalization biases in value-based learning and corresponding biases in episodic and autobiographical memory; 2) elucidate the neurocognitive mechanisms underlying these biases; 3) delineate normative longitudinal developmental changes in these processes from childhood to adulthood; and 4) establish whether computational phenotypes capturing these biases predict anxious and depressive symptomatology.