The laminar organization of 'index' versus 'attribute' coding in neocortex

新皮质中“索引”与“属性”编码的层状组织

基本信息

批准号：
10205913
负责人：
BRUCE L MCNAUGHTON
金额：
$ 198.83万
依托单位：
UNIVERSITY OF CALIFORNIA-IRVINE
依托单位国家：
美国
项目类别：
财政年份：
2021
资助国家：
美国
起止时间：
2021-04-15 至 2024-03-31
项目状态：
已结题

来源：
https://reporter.nih.gov/project-details/10205913
关键词：
3-Dimensional Adaptive Behaviors Aging Behavior Body Weight Changes Brain Categories Cells Characteristics Clinical assessments Code Cognition Cognition Disorders Cognitive deficits Complex Coupling Data Disease Electrophysiology (science)Episodic memory Event Exhibits Fright Goals Hippocampal Formation Hippocampus (Brain)Hunger Image Intelligence Intervention Knowledge Lead Lesion Light Link Location Memory Memory Disorders Memory impairment Motor Mus Neocortex Neurons Output Pattern Process Property Recurrence Rest Retrieval Scheme Schizophrenia Sensory Silicon Sleep Structure Supporting Cell Synapses Testing Trauma Weight Work artificial neural network association cortex autism spectrum disorder base density detector experience experimental study improved indexing information processing learning network memory acquisition memory recall neural circuit relating to nervous system sensory discrimination sensory input statistics theories trend two-photon

项目摘要

We propose a circuit-level principal underlying how brains acquire 'episodic' memories and reprocess them into compact, efficient 'schemas': The attributes or 'contents' of experience are represented primarily in the deeper layers of neocortex (NC), whereas the superficial layers are dedicated to encoding the contexts in which the attributes occur. Synaptic associations between superficial context codes and deep attribute codes permit contexts to evoke appropriate attribute output hence enabling memory recall and predictive behavior. The hippocampus (HC) is essential for acquisition of memories and for their reprocessing into efficient, schematic representations of the world. NC exhibits dense local but sparse long-range connectivity, which severely limits its ability to make rapid, long-range associations. HC likely solves this dilemma by merging the totality of the brain's current internal state (i.e., sensory input and internal variables such as hunger, fear, current goals) into a unique, 'index' code that is projected to NC, and associated with its current, distributed, attribute representation. Retrieval of an index code evokes the corresponding attributes. Such HC-orchestrated retrieval may enable the gradual rewiring of NC circuitry in a manner that captures the overall statistics of experience, much the same way as deep, artificial, neural networks learn incrementally by small connection weight adjustments directed by the overall statistics of the input. Our hypothesis on the laminar division of labor in this process is based on the facts that HC output is directed primarily to upper layers of NC, which implements a 'spatial' coding scheme that is lost after HC lesions; and that the deeper layers of NC frequently exhibit more robust responsiveness to and discrimination of sensory inputs than the superficial ones. We propose to record cellular level, neural ensemble activity simultaneously from deep and superficial layers in primary and association cortex, using high-density, electrophysiological recording. First, we attempt to establish the 'attribute vs index' principal by showing that deep cells shift their firing locations with shifts in the relevant sensory attributes, whereas superficial cells do not. Next we test the hypothesis that, as NC accumulates large amounts of diverse experience, attribute representations in deeper layers becomes sparser and more categorically organized, whereas superficial layer coding is relatively unchanged. To accomplish this, we employ a recent chemogenetic advance that enables us to acquire large amounts of resting-state cellular data, in which we expect the predicted changes will be most easily observed. We also explore the statistics of excitatory-inhibitory cell functional connectivity that may underlie such coding statistics changes. The expected advances in understanding cortical memory and schema encoding circuits will ultimately improve clinical assessment of, and intervention in memory and cognitive disorders.

我们提出了一个电路级原理，作为大脑如何获取“情景”记忆和重新处理的基础将它们转化为紧凑、高效的“模式”：经验的属性或“内容”主要表现为新皮质 (NC) 的深层，而表层则致力于编码上下文属性出现的位置。浅层上下文代码和深层属性代码之间的突触关联允许上下文唤起适当的属性输出，从而实现记忆回忆和预测行为。海马体 (HC) 对于记忆的获取以及将其重新加工成有效的、世界的示意图。 NC 表现出密集的本地连接但稀疏的远程连接，这严重限制了其进行快速、远程关联的能力。 HC 可能会通过合并来解决这个困境大脑当前内部状态的整体（即感觉输入和内部变量，如饥饿、恐惧、当前目标）转换成一个独特的“索引”代码，该代码被投影到 NC，并与其当前的、分布式的、属性表示。检索索引代码会唤起相应的属性。这样的HC精心策划检索可以以捕获总体统计数据的方式逐步重新布线 NC 电路经验，与深度人工神经网络通过小连接增量学习的方式非常相似由输入的总体统计数据指导权重调整。我们关于层层划分的假设这一过程中的劳动力基于以下事实：HC 输出主要针对 NC 上层，这实施 HC 损伤后丢失的“空间”编码方案；并且 NC 的更深层次经常与表面的感觉输入相比，对感觉输入表现出更强大的反应和辨别能力。我们建议从深层和表层同时记录细胞水平、神经整体活动使用高密度电生理记录来分析初级皮层和联合皮层的各层。首先，我们尝试通过表明深层细胞随着神经元的变化而改变其发射位置，建立了“属性与索引”原理相关的感觉属性，而表面细胞则没有。接下来我们检验假设，作为 NC 积累大量不同的经验，更深层的属性表示变得更稀疏并且组织更加明确，而表层编码相对没有变化。为了实现这一目标，我们采用了最近的化学遗传学进展，使我们能够获得大量静息态细胞数据，我们期望在其中最容易观察到预测的变化。我们还探索了统计数据兴奋-抑制细胞功能连接可能是这种编码统计变化的基础。预期的理解皮质记忆和图式编码电路的进步将最终改善临床记忆和认知障碍的评估和干预。

项目成果

期刊论文数量（4）

专著数量（0）

科研奖励数量（0）

会议论文数量（0）

专利数量（0）

Replay, the default mode network and the cascaded memory systems model.

重放、默认网络模式和级联存储系统模式。

DOI：
发表时间：
2022-10
期刊：
影响因子：
0
作者：
Kaefer, Karola;Stella, Federico;McNaughton, Bruce L;Battaglia, Francesco P
通讯作者：
Battaglia, Francesco P

Cortical reactivation of spatial and non-spatial features coordinates with hippocampus to form a memory dialogue.

空间和非空间特征的皮层重新激活与海马体协调形成记忆对话。

DOI：
发表时间：
2023-11-27
期刊：
影响因子：
16.6
作者：
Chang, HaoRan;Esteves, Ingrid M;Neumann, Adam R;Mohajerani, Majid H;McNaughton, Bruce L
通讯作者：
McNaughton, Bruce L

Coordinated activities of retrosplenial ensembles during resting-state encode spatial landmarks.

静息状态下压后群的协调活动编码空间标志。

DOI：
发表时间：
2020-05-25
期刊：
影响因子：
0
作者：
Chang, HaoRan;Esteves, Ingrid M;Neumann, Adam R;Sun, Jianjun;Mohajerani, Majid H;McNaughton, Bruce L
通讯作者：
McNaughton, Bruce L

Learning in deep neural networks and brains with similarity-weighted interleaved learning.

通过相似性加权交错学习在深度神经网络和大脑中进行学习。