Spatiotemporal neuronal system dynamics underlying hierarchical visual representations of objects and faces for primate perception and discrimination

人类和（非人类灵长类）的大脑是最复杂的生物系统之一。神经科学的一个主要挑战是理解大脑作为一个动态复杂系统是如何运作的，以及正常感知和认知背后的神经元机制是什么。物体和人脸的视觉表征在多个相互连接的区域同时编码，具有高度并行化和分层组织的处理阶段。我们的目标是通过发现在颞叶和与选择相关的前额叶区域的高级视觉区域内部和之间运行的时空过程来推进对系统级神经元相互作用的科学理解，这些过程共同构成了物体/面部感知和歧视的基础。我们只能通过结合多电极、多区域记录和干预来了解大脑区域在神经元水平上的因果关系。要做到这一点，我们必须记录大脑的基本功能单位，神经元，（而不是包含数十万个神经元的神经成像“体素”），我们必须使用动物模型，因为这样的记录是侵入性的。最近的技术进步促进了多区域多电极记录和研究许多区域和皮层内部和之间的神经元动力学。此外，我们只能通过将神经元记录与干预相结合来了解大脑区域如何在神经元水平上因果相互作用。自Hebb的突触可塑性原理以来，相互连接的神经元活动的时空动态被认为是学习的关键原理。时空放电模式存在于不同行为任务的许多区域。在视觉的情况下，分布式视觉表征的结合可以利用这些原理，但在多个时间尺度和时间滞后下提取复杂的神经元放电（“spike”）序列在计算上是复杂的。现在，新的高效算法已经开发出来，用于识别在不同时间尺度上具有一致峰值延迟的较大神经元集合，而无需假设底层编码方法。我们将应用新的计算和统计工具来识别细胞组合，并提取区域内和区域间一致的多神经元尖峰序列。我们的经验记录将通过运行GPU优化的尖峰神经网络模型模拟来补充，以评估我们的发现并生成新的预测。同时，更多神经元的记录提高了“维数诅咒”和网络科学，为系统神经科学提供了先进的分析和可扩展工具，补充了降维方法，并允许对网络结构进行定量分析和对人口范围内活动的概率描述。这些方法使我们能够更好地理解计算，量化和跟踪关键概念（如“细胞组装”）的动态，并跟踪由行为或大脑干预引起的变化。我们的目标分为3个子目标：前两个目标是了解多个高级视觉区域和皮层内部和之间的动态时空表征和神经元相互作用机制，这些区域和皮层分别构成了对物体和面部的正常感知。就面部而言，我们将研究不同颞叶和额叶面部斑块之间的动态和相互作用。我们的第三个子目标是了解这些机制和相互作用在记忆、分类和选择行为方面对物体和面孔的不同，特别强调额叶-颞叶的相互作用。

The human and (non-human primate) brain is one of the most complex biological systems. A major challenge in neuroscience is to understand how the brain operates as a dynamic complex system and what neuronal mechanisms underlie normal perception and cognition. Visual representations of objects and faces are simultaneously encoded across multiple reciprocally connected regions with highly parallelized and hierarchically organized processing stages. Our objective is to advance scientific understanding of systems level neuronal interactions by discovering the spatio-temporal processes operating within and between higher visual areas in the temporal lobe and choice-related prefrontal regions, that together underlie object/face perception and discrimination. We can only learn how brain areas causally interact at the neuronal level by combining multi-electrode, multi-area recordings with interventions. To do this we must record the fundamental functional units in the brain, neurons, (not neuroimaging 'voxels' containing hundreds of thousands of neurons) and we must use animal models because such recording is invasive. Recent technological advances facilitate multi-area multi-electrode recordings and investigation of neuronal dynamics both within and between many areas and cortical layers. Moreover, we can only learn how brain areas causally interact at the neuronal level by combining neuronal recordings with interventions. Ever since Hebb's principles of synaptic plasticity, spatiotemporal dynamics of interconnected neuronal activities have been implicated as key principle underlying learning. Spatiotemporal firing patterns exist in many areas across different behavioral tasks. In the case of vision, the binding of distributed visual representations may exploit such principles but the extraction of complex neuronal firing ('spike') sequences, at multiple temporal scales and time-lags, is computationally complex. Now, new eEfficient algorithms have been developed for identifying larger assemblies of neurons with consistent spike delays at varying temporal scales without making assumptions of the underling encoding method. We will apply new computational and statistical tools to identify cell assemblies and extract consistent multi-neuron spiking sequences within and across areas. Our empirical recordings will be complemented by running GPU optimised simulations of spiking neural networks models to assess our findings and generate novel predictions. Recording from more neurons simultaneously raises the 'curse of dimensionality' and Network science, offers advanced analytical and scalable tools for systems neuroscience, complementing dimension-reduction approaches, and allowing for quantitative analyses of network structure and probabilistic descriptions of population-wide activity. These approaches allow us to better understand computations, quantify and track dynamics of key concepts such as 'cell assemblies', and track changes elicited by behaviour or brain interventions.Our objective is divided into 3 sub-goals: The first two are to understand the dynamic spatiotemporal representations and mechanisms of neuronal interaction operating within and between multiple higher visual areas, and cortical layers, that underlie normal perception of objects and faces respectively. In the case of faces we will study such dynamics and interactions across different temporal and frontal lobe face patches. Our third sub-goal is to understand how these mechanisms and interactions differ in the context of memory, categorization, and choice behaviour with respect to objects and faces with a special emphasis on frontal lobe - temporal lobe interactions.