Neural mechanisms for reducing interference during episodic memory formation

描述（由申请人提供）：成功记忆的最大挑战之一是相似记忆之间的混淆或干扰。对于我们存储在记忆中的每一个密码、名字或停车位，都有许多其他的密码、名字或停车位是我们已经学会或将来会学会的。虽然干扰是“正常”遗忘的一个相对良性但令人烦恼的因素，但它也是伴随衰老和/或痴呆而发生的临床遗忘的一个主要因素。因此，有一个基本的需要，了解神经机制，支持获取/检索类似的记忆，同时尽量减少干扰和相应的遗忘。情景记忆的计算模型提出了两种被认为可以减少干扰相关遗忘的核心机制：整合和模式分离。整合包括将重叠的记忆“融合”到一个共同的表象中，这样这些记忆之间的关系更像是互补而不是竞争。模式分离涉及到相似记忆的正交化，这样记忆之间的差异被夸大，干扰的可能性被最小化。虽然人们普遍认为这些机制在理论上很有吸引力，并提供了明确的计算优势，但关于如何以及何时调用这些学习机制的明确证据——尤其是在人类身上——仍然令人惊讶地有限。特别是，在(a)每种机制可能被招募的学习环境中，(b)每种机制对应的神经特征是什么，以及(c)与每种机制的参与相关的具体行为后果方面，仍然存在模糊性。我们建议对整合和模式分离发生的背景进行系统的调查，目的是使用复杂的、尖端的神经成像（fMRI）技术来识别可诊断每种机制的神经活动的分布模式。重要的是，我们还计划利用这些观察到的神经活动模式——即整合与分离的神经证据——来预测行为记忆现象，包括与干扰相关的遗忘。该研究是心理学和神经科学问题的强大综合，重点是受机器学习和数据挖掘领域的计算模型和分析方法启发的学习机制。

DESCRIPTION (provided by applicant): One of the biggest challenges to successful remembering is the potential for confusion or interference between similar memories. For every password, name, or parking space that we store in memory, there are many other passwords, names or parking spaces that we have already learned or will learn in the future. While interference is a factor in relatively benign-yet annoying-examples of 'normal' forgetting, it is aso a major factor in clinically significant examples of forgetting that occur with aging and/or dementia. Thus, there is a fundamental need to understand the neural mechanisms that support the acquisition/retrieval of similar memories while minimizing interference and corresponding forgetting. Computational models of episodic memory have proposed two core mechanisms that are thought to reduce interference-related forgetting: integration and pattern separation. Integration involves 'fusing' overlapping memories into a common representation such that the relationship between these memories is more complementary than competitive. Pattern separation involves the orthogonalization of similar memories such that differences between memories are exaggerated and the potential for interference minimized. While there is general agreement that these mechanisms are theoretically appealing and offer clear computational advantages, clear evidence for how and when these learning mechanisms are invoked- particularly in humans-remains surprisingly limited? In particular, there remains ambiguity as far as (a) the learning contexts in which each mechanism might be recruited, (b) what the corresponding neural signatures of each mechanism are, and (c) the specific behavioral consequences associated with the engagement of each mechanism. We propose a systematic investigation of the contexts in which integration and pattern separation occur with the goal of using sophisticated, cutting-edge neuroimaging (fMRI) techniques to identify distributed patterns of neural activity that are diagnostic of each mechanism. Critically, we also plan to use these observed patterns of neural activity-that is, neural evidence for integration vs. separation-to predict behavioral memory phenomena, including interference-related forgetting. The research represents a strong synthesis of psychology and neuroscience questions with an emphasis on learning mechanisms inspired by computational models and analysis approaches that draw from the fields of machine learning and data mining.