The computational and neural mechanisms linking decision-making and memory in humans

项目概要 决策如何影响记忆，从而对人类行为产生长期影响？回答这个 这个问题对于理解受损的决策如何可能导致适应不良的记忆结果至关重要， 例如沉思负面事件、容易产生错误记忆或记忆丧失。近年来， 强化学习（RL）框架对于描述受损决策尤其有效 精神疾病以及识别连接决策和记忆的计算机制。的 该项目的首要目标是确定神经计算机制，以解释学习如何进行 驱动决策的过程也会影响随后的记忆。为此，建议的 K99 阶段 研究包括一种计算方法来确定无模型强化学习信号如何影响 人类海马和非海马识别记忆表现（目标 1），以及神经生物学- 确定这些 RL-记忆相互作用背后的电生理机制的方法（目标 2）； 研究的 R00 阶段将部署这些方法来研究基于模型的强化的贡献 学习这些不同的记忆过程（目标 3）。具体来说，目标 1 将测试无模型 RL 如何发出此类信号 因为预测错误可能会与刺激的感知特征相互作用，从而增强即时（非 健康志愿者的海马）记忆和延迟（海马）记忆。目标 2 将利用颅内 从带有癫痫监测电极的人类身上获得的记录，以测试额叶皮质中的神经活动如何 tex、海马和非海马内侧颞区（例如海马旁回）辅助 无模型强化学习过程对记忆性能的影响。完成目标 1-2 (K99) 后， 候选人——具有人类记忆神经生物学背景的神经科学家——将接受新的培训 在强化学习和决策的计算建模方面拥有独特的、多学科的 适用于目标 3 (R00) 的技能组，研究基于模型的 RL 过程如何影响助记行为和神经网络 活动。鉴于候选人的导师的专业知识，他们非常适合为这些目标提供培训 桥接计算模型（顾晓思博士）和人类神经生理学（Ignacio Saez 博士）以理解 人类决策。这些技能还将促进候选人转变为独立研究员 长期目标是进行综合行为、计算和生物学研究，了解这些 人类的决策和记忆过程在精神疾病中出现错误。

PROJECT SUMMARY How does decision-making influence memory to leave a long-lasting impact on human behavior? Answering this question is critical to understanding how impaired decision making might lead to maladaptive memory outcomes, such as rumination on negative events, susceptibility to false memories, or memory loss. In recent years, the reinforcement learning (RL) framework has been particularly fruitful for describing impaired decision-making in psychiatric disorders as well as identifying computational mechanisms linking decision-making and memory. The overarching aim of this project is to identify the neurocomputational mechanisms that explain how the learning processes driving decision-making also influence subsequent memory. To do so, the K99 phase of the proposed study consists of a computational approach to identify how model-free reinforcement learning signals influence both hippocampal and non-hippocampal recognition memory performance in humans (Aim 1), and a neurobiolog- ical approach to identify the electrophysiological mechanisms underlying these RL-memory interactions (Aim 2); the R00 phase of the study will deploy these approaches to study the contributions of model-based reinforcement learning to these distinct memory processes (Aim 3). Specifically, Aim 1 will test how model-free RL signals such as prediction errors might interact with the perceptual features of a stimulus to enhance both immediate (non- hippocampal) memory and delayed (hippocampal) memory in healthy volunteers. Aim 2 will leverage intracranial recording obtained from humans with epilepsy monitoring electrodes to test how neural activity in the frontal cor- tex, hippocampus, and non-hippocampal medial temporal regions (such as parahippocampal gyrus) subserve the influence of model-free RL processes on memory performance. Upon completion of Aims 1-2 (K99), the candidate—a neuroscientist with a background in the neurobiology of human memory—will obtain new training in computational modeling of reinforcement learning and decision-making and have a unique, multi-disciplinary skillset to apply to Aim 3 (R00), to study how model-based RL processes influence mnemonic behavior and neural activity. The candidate’s mentors are uniquely suited to provide the training for these Aims, given their expertise in bridging computational modeling (Dr. Xiaosi Gu) and human neurophysiology (Dr. Ignacio Saez) to understand human decision-making. These skills will also facilitate the candidate’s transition into an independent researcher with the long-term goal of performing integrative behavioral, computational, and biological studies of how these human decision-making and memory processes go awry in psychiatric disorders.