Modeling Face Perception

描述(申请人提供)：在理解复杂的视觉对象，特别是人脸是如何被大脑皮层处理的方面，已经有了很大的进展。与此同时，复杂的神经网络模型已经被开发出来，它们可以完成这些大脑皮层区域所需的许多相同任务。这项建议的目的是扩展这些简化的模型，以了解面部表情处理在多大程度上是“普遍的”，而不是它在多大程度上是经验调节的。特别是，我们通过建模重点阐明了以下问题：(1)理解表情识别中的文化差异：对比其他种族效应和文化展示规则的影响；(2)理解我们是如何成为面子“专家”的。为什么像我们擅长的其他视觉任务一样，梭形回的同一区域也会被招募到面部？(3)理解面部专业知识的动态：如何规划眼睛运动来有效地提取特征？在每一种情况下，我们都已经开发了或将开发出该过程的神经计算模型。文化和其他种族的体验将由我们模型的内部表征和训练信号的组成来建模。面部专业知识被建模为不同任务要求(需要不同的分类级别)和培训时间的组合。眼动建模将基于新的和传统的方法来提取每个任务的面部信息位置。我们将开发一个基于特征值和任务所需类别之间的互信息的眼跳目标人脸的理论标准。我们还将进行行为实验来测试我们模型的预测。该项目将阐明大脑代表和处理面孔的方式，并将深入了解面孔处理缺陷的潜在问题，如面孔失认症。

DESCRIPTION (provided by applicant): There has been a great deal of progress in understanding how complex visual objects, in particular, human faces, are processed by the cortex. At the same time, sophisticated neural network models have been developed that do many of the same tasks required by these cortical areas. The aim of this proposal is to extend these simplifying models toward an understanding of the extent to which facial expression processing is "universal," versus the extent to which it is experientially mediated. In particular, we focus upon elucidating the following issues through modeling: (1) Understanding cultural variation in expression recognition: contrasting the influence of other race effects versus cultural display rules; (2) Understanding how we become face "experts." Why does the same region of the Fusiform Gyrus get recruited for faces as well as other visual tasks we may be expert in? (3) Understanding the dynamics of facial expertise: How are eye movements planned for efficient feature extraction? In each case, we have developed or will develop a neurocomputational model of the process. Cultural and other-race experience will be modeled by the composition of the internal representations and the training signals of our model. Facial expertise is modeled as a combination of different task requirements (varying the level of categorization required) and length of training. Eye movement modeling will be based on novel and traditional methods for extracting the informative locations on the face for each task. We will develop a theoretical criterion for saccade targets the face based upon mutual information between feature values and the categories required for the task. We also will be performing behavioral experiments to test the predictions of our model. The project will shed light on the way faces are represented and processed by the brain, and should give insights into the problems underlying deficits in face processing such as prosopagnosia.