Integrating and separating information sequences in the human cerebral cortex

项目摘要 认知的许多方面都会随着时间的推移而延续。听到一段声音，我们将其视为 知更鸟的旋律；听到一个单词，我们就把它理解为一个有意义的句子的一部分。因此，我们的 随着时间的推移，大脑必须具备整合信息的能力。但是，信息不能被集成 不加区别地：新句子的主语不一定与前一句的动词相关， 我们可能想要将这些信息项分开。这项工作的目标是理解 我们的大脑在分离无关信息的同时，灵活地整合相关信息的算法。在……里面 此外，我们的目标是了解时间整合和分离是如何在大脑皮层电路中实现的，以及 我们如何才能操纵这些大脑过程。我们之前的工作表明，大脑在一种 分布方式：随着时间的推移，人类大脑皮层的几乎所有区域都可以整合信息。早期感觉 区域在短时间内整合(毫秒到秒)并且它们将信息传递到更高阶的区域， 它们在更长的时间段(几秒到几分钟)内集成。针对这一过程，我们提出了以下算法： 每个皮质区域都保持着自己的局部记忆，并试图形成一个合成的联合表示 它的本地存储器和它接收到的任何输入。当这种合成成功时，大脑皮层区域将通过 将其合成表示转发到下一处理阶段。但如果合成不成功，那么 它的本地内存将被重置。例如，早期皮层区域可能会合成单词内的音节，但 然后在一个新词的开头重置其上下文。为了测试大脑是否正在使用这种算法，我们将 在人们的大脑中模拟fMRI活动，这些人在听的时候整合和分离信息序列 到复杂的叙事。了解如何在活动中实现时态集成和分离 对于大脑皮层电路，我们还将测量ECoG信号，它提供同步和 大脑皮层神经元的非同步活动。我们假设大脑皮层回路在以下情况下变得不那么同步 他们不能将新的输入与先前的背景相结合。我们将测试这种低频的下降是否 同步允许更多的信息从世界流向大脑皮层层次。最后，我们会 开发工具来控制时间整合或分离的状态，使用电气调制来增加或 降低大脑皮层回路的同步性。总而言之，这项工作提供了检测和处理 人脑的时间整合和分离过程，以及一个计算框架 我们如何分析信息序列。我们的方法是创新的，因为我们开发了多尺度 实验范例，并将fMRI、ECoG和建模的数据结合在一起。我们希望这项工作有助于揭示 人类大脑皮层电路整合和分离信息序列的层级机制。

Project Summary Many aspects of cognition extend over time. Hearing a fragment of sound, we perceive it as part of a mockingbird’s melody; hearing one word, we understand it as part of a meaningful sentence. Therefore, our brains must possess the ability to integrate information over time. However, information cannot be integrated indiscriminately: the subject of a new sentence is not necessarily related to the verb of the previous sentence, and we may want to keep these items of information separate. The goal of this work is to understand the algorithms by which our brains flexibly integrate related information while separating unrelated information. In addition, we aim to understand how temporal integration and separation are implemented in cortical circuits, and how we can manipulate these brain processes. Our prior work suggests that brains perform this task in a distributed manner: almost all regions of the human cortex can integrate information over time. Early sensory regions integrate over short periods (milliseconds to seconds) and they pass information to higher-order regions, which integrate over longer periods (seconds to minutes). We propose the following algorithm for this process: each cortical region maintains its own local memory, and attempts to form a synthesized joint representation of its local memory and any input that it receives. When this synthesis is successful, a cortical region will pass forward its synthesized representation to the next stage of processing. But if the synthesis is unsuccessful, then its local memory will be reset. For example, an early cortical region may synthesize syllables within a word but then reset its context at the beginning of a new word. To test whether the brain is using this algorithm, we will model fMRI activity in the brains of people who integrate and separate sequences of information as they listen to complex narratives. To understand how temporal integration and separation are implemented in the activity of cortical circuits, we will also measure ECoG signals, which provide a direct read-out of synchronous and asynchronous activity in cortical neurons. We hypothesize that cortical circuits become less synchronized when they cannot integrate new input with prior context. We will test whether this decrease in low-frequency synchronization allows increased information flow from the world into the cortical hierarchy. Finally, we will develop tools to control the state of temporal integration or separation, using electrical modulation to increase or decrease synchronization in cortical circuits. Altogether, this work provides tools to detect and manipulate temporal integration and separation processes in the human brain, as well as a computational framework for how we parse sequences of information. Our approach is innovative because we develop multi-scale experimental paradigms and combine data across fMRI, ECoG, and modeling. We expect this work to help reveal the hierarchical mechanisms by which human cortical circuits integrate and separate sequences of information.