Proj 3: Retrieval Dynamics in Item and Source Memory (p. 207 - 236)

这项研究的目的是阐明记忆提取的机制，它们如何随着时间的推移而展开， 它们如何受到自上而下的影响，如受试者的目标或偏见。这些问题 在项目识别任务的背景下进行研究，受试者判断测试项目是否是 研究列表，以及来源回忆任务，其中被试判断项目是否具有一定的上下文 美德.先知-愿关于受试者的决策、反应时间和大脑活动测量的实验数据 在提取过程中，用于测试和促进记忆的计算模型的发展 检索动力学第一个具体的目标是联合收割机项目识别的总和相似性模型（即 指定记忆信号的强度）与漂移扩散模型（指定检索动态，给定 信号强度）。将通过将组合模型拟合到两个行为数据（准确性）来测试组合模型。 和RT分布）和电生理数据（头皮和颅内EEG中观察到的振荡效应 录音）。第二个具体目标是发展漂移扩散模型的扩展，以解释 自上而下的控制如何影响源存储器性能。扩展模型将用于推导 检索应该如何定向（即，受试者是否应该尝试从一个来源或另一个来源检索记忆） 以最大化正确的性能。模型预测将通过将模型拟合到行为数据来进行测试， 还通过使用模式分类算法（应用于fMRI数据）来提取以下信息的“神经指数”： 检索取向在模型参数和神经数据之间建立更好的联系将大大提高我们的 从行为中预测神经活动的能力，反之亦然。此外，这项研究将有利于 诊断和治疗患者的记忆障碍;通过连接特定模式的神经和 具有特定参数的行为数据，该模型可以指定（精确，定量 术语），其中记忆决策过程在给定患者中出错。

The goal of this research is to elucidate the mechanisms of memory retrieval, how they unfold in time, and how they are modulated by top-down influences, such as subjects' goals or biases. These problems are studied in the context of an item recognition task, where subjects judge whether a test item was part of a studied list, and a source recall task, where subjectsjudge whether items possess certain contextual attributes. Experimental data on subjects' decisions, their reaction time, and measures of their brain activity during the retrieval process are used to test and facilitate development of a computational model of memory retrieval dynamics. The first specific aim is to combine summed-similarity models of item recognition (that specify the strength of the memory signal) with drift-diffusion models (that specify retrieval dynamics, given a particular signal strength). The combined model will be tested by fitting it to both behavioral data (accuracy and RT distributions) and electrophysiological data (oscillatory effects seen in scalp and intracranial EEG recordings). The second specific aim is to develop an extension of the drift-diffusion model to account for how top-down control influences source memory performance. The extended model will be used to derive how retrieval should be oriented (i.e., should subjects try to retrieve memories from one source or the other) to maximize correct performance. Model predictions will be tested by fitting the model to behavioral data, and also by using pattern classification algorithms (applied to fMRI data) to extract a "neural index" of retrieval orientation. Forging better links between model parameters and neural data will greatly improve our ability to predict neural activity from behavior, and vice-versa. Furthermore, this research will benefit diagnosis and treatment of patients with memory disorders; by linking specific patterns of neural and behavioral data with specific parameters, the model makes it possible to specify (in precise, quantitative terms) where the memory decision-making process is going awry in a given patient.