Learning to Represent Space in the Brain

我们日常生活中的许多行为--上班、探索城市或在房间里走动--都需要导航。为了导航，我们需要感知我们所在的位置，为此我们主要使用视觉。这项提议的目标是了解大脑回路是如何学习使用视觉信息来支持导航的。众所周知，大脑的导航区域代表空间，包括海马体的位置细胞和内嗅觉皮质的网格细胞。视觉皮质区域的输入支持对空间的这种表示。处理空间表示的传统模型是分层模型，其中视觉皮层区域在视觉皮质层次上构建越来越精细的视觉特征表示，然后将这些表示馈送到导航区域。然而，最近的结果发现，即使在初级视觉皮质中也有空间的表征，这表明了一种更分散的处理模式。这导致了一个谜团，即哪种相互冲突的模式是正确的：层级模式还是分布式模式。基于视觉皮层中空间信号的逐渐出现，我假设它们有可能在不同的时间为真，而学习将表征从层次化转变为分布式。我们将结合虚拟现实、来自多个大脑区域的大规模录音，以及从特定反馈投射到视觉皮质的失活和录音来验证这一中心假设。导航是所有动物共有的一种基本认知能力，是一种融合了感知、记忆、规划和决策的认知原型。因此，了解空间表征是如何在大脑中构建的，为理解导航本身和一般认知能力的更广泛机制提供了机会。因此，它具有改变感觉神经科学、记忆研究和神经形态机器学习领域的潜力。

Many actions in our daily lives - getting to work, exploring a city or moving around a room - involve navigation. To navigate we need to sense where we are, for which we predominantly use vision. The goal of this proposal is to understand how brain circuits learn to use visual information to support navigation.Navigational areas of brain are known to represent space, including place cells of the hippocampus and grid cells of the entorhinal cortex. This representation of space is supported by inputs from visual cortical areas. The conventional model for processing a representation of space is a hierarchical model, where visual cortical areas build increasingly elaborate representations of visual features over the visual cortical hierarchy, and then feed these representations to navigational areas. Recent results however find a representation of space even in the primary visual cortex, suggesting a more distributed model of processing. This leads to a mystery as to which of the conflicting models is true: hierarchical or distributed. Based on the gradual appearance of spatial signals in the visual cortex, I hypothesize that it is possible for them to be true at different times, and learning transforms the representation from hierarchical to distributed. We will test this central hypothesis using a combination of virtual reality, large scale recordings from multiple brain regions, and inactivations and recordings from specific feedback projections to the visual cortex. Navigation is an essential cognitive ability common to all animals, and is a cognitive archetype, fusing perception, memory, planning and decision making. Understanding how spatial representations are built in the brain therefore offers the opportunity to understand both navigation itself and the wider mechanisms of general cognitive abilities. It thus has the potential to transform the fields of sensory neuroscience, memory research and neuromorphic machine learning.