Neural Mechanisms of Face Recognition

描述(申请人提供)：对人脸识别的理解存在一个根本性的差距：为什么大脑中有多个针对人脸的神经元代码，以及它们是如何通过相互连接的人脸处理区域网络进行转换以支持人脸识别的？这一差距的持续存在构成了一个重要问题，因为人脸具有社会重要性，而且在填补这一差距之前，物体识别的神经机制在很大程度上仍然无法理解。这项拟议研究的长期目标是在高度进化的人脸处理系统中获得对人脸识别的机械理解，类似于人类的人脸处理系统。这一特定应用的总体目标是识别面部处理网络的皮质节点之间的信息神经元转换的规则和机制。这项拟议的工作将检验中心假设，即这个网络被组织为一个信息处理层次，服务于稳健的人脸识别，其中每个处理级别执行特定于人脸的转换。它利用了模型系统的功能组织，该组织由空间上不同但相互关联的 这些节点具有独特的功能专门化，而且由于它们对已知的视觉对象类别-面部的选择性，这些节点很容易通过脑成像进行识别。 这一建议的基本原理是，在完成这项研究后，我们将在计算、表征和机械水平上了解高级对象识别的核心操作。在强大的初步数据的指导下，中心假设将通过追求三个具体目标来验证：1)人脸细胞使用什么视觉特征来表征复杂的人脸信息？2)人脸区域如何相互作用来生成高维人脸编码？3)人脸区域在人脸编码和人脸检测中的因果作用是什么？在第一个目标下，我们将结合大脑成像识别的三个面部区域的单个单位电生理记录和参数视觉刺激，以揭示单个细胞用于编码面部信息的计算机制。在第二个目标下，将分析来自多个区域的联合电生理记录，以揭示区域间的相互作用如何产生不同区域和随时间变化的面部表征。在目标3下，将使用定向失活来揭示不同的人脸处理区域在信息转换和人脸检测中所起的因果作用。这项拟议的研究意义重大，因为它有望直接展示信息处理网络是如何组织起来的，以转换高级对象类别的视觉表示，并将它们用于视觉行为。这样做将把我们对视觉对象识别的理解提升到一个新的水平。提出的这项研究在概念和方法上都是创新的，因为它从系统的角度看待目标识别问题，通过多节点网络跟踪信息的转换，并通过一种新的方法组合，将单细胞机制和种群代码的分析与网络功能的因果询问结合在一起。

DESCRIPTION (provided by applicant): There is a fundamental gap in understanding of face recognition: why are there multiple neuronal codes for faces in the brain, and how they are transformed through a network of interconnected face-processing areas to support face recognition? Continued existence of this gap constitutes an important problem because of the social importance of faces and because, until it is filled, the neural mechanisms for object recognition remain largely incomprehensible. The long-term goal of the proposed research is to gain a mechanistic understanding of face-recognition in a highly evolved face-processing system similar to that of humans. The overall objective of this particular application is to identiy the rules and mechanisms of informational neuronal transformations between the cortical nodes of the face-processing network. The proposed work will test the central hypothesis that this network is organized as an information-processing hierarchy serving robust face-recognition, in which each processing level performs a face-specific transformation. It takes advantage of the functional organization of the model system that consists of spatially distinct, but interconnected nodes with unique functional specializations - and the fact that these nodes are readily identifiable with brain imaging due to their selectivity for a known visual object category, faces. The rationale of this proposal is that, after completion of this research, we will understand core operations of high-level object recognition at a computational, representational, and mechanistic level. Guided by strong preliminary data, the central hypothesis will be tested by pursuing three specific aims: 1) What visual features do face cells use to represent complex facial information? 2) How do face areas interact to generate high-dimensional facial codes? 3) What is the causal role of face areas for facial coding and face detection? Under the first aim, we will combine single unit electrophysiological recordings in three brain-imaging identified face areas with parametric visual stimulation to reveal the computational mechanisms single cells use to code facial information. Under the second aim, joint electrophysiological recordings from multiple areas will be analyzed to reveal how inter-areal interactions generate face-representations that differ across areas and change over time. Under aim 3, targeted inactivation will be used to reveal the causal role different face processing areas play for informational transformations and for face detection. The proposed research is significant, because it is expected to directly show how an information processing network is organized to transform visual representations of a high-level object category and utilizes them for visual behavior. In doing so it will lift our understanding of visual object recognition to a new level. The research proposed is conceptually and methodologically innovative because it in order to take a systems perspective to the problem of object recognition, tracing transformations of information through a multi-node network and integrating, through a novel combination of methodologies, analyses of single cell mechanisms and population codes with causal interrogation of network function.