Dissociating spatial and cognitive grid representations in the brain

分离大脑中的空间和认知网格表征

• 批准号: 10655777
• 负责人: Matthew Nassar
• 金额: $ 16.25万
• 依托单位: BROWN UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-08-01 至 2023-07-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10655777
• 关键词: Dissociating; spatial; cognitive; grid; representations

Humans and primates display an impressive knack for acquiring knowledge in one environment (for example the arrangement of produce and dairy at the grocery store) and rapidly applying it in a completely novel one (a new grocery store). The ability to generalize behavior from past contexts to new ones is critical to prediction, inference, and planning in the real world. Though models of artificial intelligence inspired by neural learning algorithms have achieved human levels of performance on a wide range of tasks, the most prominent such models are incapable of generalizing information beyond the problems that they were trained on. In principle, such transfer requires extracting abstract information from one situation and then applying it in another, yet the neural mechanisms of this abstraction process remain unknown. Here we propose to test the hypothesis that the brain achieves such transfer using an abstract cognitive map, much like that used to navigate physical space. In particular, we propose to test whether grid-like representations in the brain, which are thought to facilitate spatial navigation and can be observed in neural firing or fMRI BOLD activity, tile an abstract space (eg. the layout of a typical grocery store) and facilitate the transfer of information collected in one environment (eg. Whole Foods) to another (eg. Eastside Market). To do so, we developed a novel task in which abstract grid representations can be dissociated through neuroimaging from more standard spatial grid representations. We will test which of these patterns is observed in the BOLD activity of the entorhinal cortex concurrent with task performance, and the degree to which such patterns relate to behavioral measures of transfer learning. We will incorporate results from these analyses into a model that learns to associate cues with locations via grid-like basis functions. The results of this study will shed light on the neural mechanisms of transfer learning and serve as preliminary data for an R01 application focused on neural representations for efficient learning.

人类和灵长类动物在一个环境中表现出令人印象深刻的获取知识的诀窍(例如 在杂货店安排农产品和乳制品)并迅速将其应用于一种全新的产品(一种新的 杂货店)。将行为从过去的上下文概括到新的上下文的能力对于预测、推理、 并在现实世界中进行规划。尽管受神经学习算法启发的人工智能模型 在各种任务上达到人类水平的表现，最突出的这类模型是无法做到的 泛化他们所训练过的问题之外的信息。原则上，这种转移需要 从一种情况中提取抽象信息，然后将其应用于另一种情况，但这一过程的神经机制 抽象过程仍不得而知。 在这里，我们建议使用抽象的认知图来测试大脑实现这种转移的假设，MART 就像用来导航物理空间的那种。特别是，我们建议测试 大脑，被认为有助于空间导航，可以在神经放电或fMRI大胆活动中观察到， 把抽象的空间铺成瓷砖(例如典型杂货店的布局)，并促进在 一种环境(例如Whole Foods)出售给另一家公司(例如东区街市)。为了做到这一点，我们开发了一个新的任务， 抽象网格表示可以通过神经成像从更标准的空间网格中分离出来 申述。我们将测试在内嗅皮层的大胆活动中观察到的这些模式中的哪一种 与任务绩效并发，以及这些模式与迁移的行为测量相关的程度 学习。我们将把这些分析的结果合并到一个模型中，该模型通过以下方式学习将线索与位置关联 网格型基函数。这项研究的结果将有助于揭示迁移学习的神经机制和 作为R01应用程序的初步数据，专注于高效学习的神经表示法。