CAREER: The Tuning and Topography of the Ventral Visual Stream

人类可以识别成千上万的物体，比如杯子、汽车、鞋子和建筑物，所有这些都是通过进入我们眼睛的光线模式来识别的。一个重要的科学前沿是了解大脑是如何完成这一壮举的。入射的光波是如何被大脑转换成我们周围世界物体的表征的？这个项目的目标是描述整个视觉系统中物体信息的组织特征——在视觉处理的不同阶段，什么样的视觉信息被表示出来，以及它是如何沿着大脑表面映射的？通过使用多种方法，包括对人类大脑的测量，再加上计算建模方法，该项目将促进我们对视觉大脑组织的理解。该项目还将促进女性在科学领域的进步，突出计算领域的女性，目标是将年轻女性的科学事业培养成学术教授和以研究为重点的行业职位。第一个目标将集中在沿着视觉处理层次调整大脑对物体反应的本质。具体来说，该项目将检验一个假设，即与一般纹理和形状信息相关的中级特征与物体的高级属性（如它们的动画性和现实世界的大小）紧密相关。第二个目标将通过使用行为和深度神经网络测量视觉和认知相似性空间，然后将这些与功能神经成像测量的神经反应联系起来，表征腹侧流反应的调节。第三个目标将引入一个计算框架，使用Kohonen自组织地图将调谐与地形联系起来。具体而言，PI将首先关注早期视觉皮层视网膜切除的大规模组织作为概念证明，然后将该框架扩展到对象响应皮层。因此，PI的目标是将空间组织的数学模型与视觉特征的深度神经网络表示相结合，以产生与人类大脑皮层组织直接相比较的人工皮层地图。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Humans can recognize thousands of objects, like cups, cars, shoes, and buildings, all from patterns of light entering our eyes. A major scientific frontier is to understand how the brain accomplishes this feat. How are incoming light waves converted by the brain into representations of objects in the world around us? The goal of this project is to characterize the organization of object information across the visual system--what kind of visual information is being represented at different stages of the visual processing, and how is it mapped along the brain’s surface? By using a combination of methods, including measurements of human brains, coupled with computational modeling approaches, this project will advance our understanding of visual brain organization. The project will also promote women's advancement in science, highlighting women in computational fields, with the goal to foster young women's scientific careers into academic professorships and research-focused industry positions.The first aim will focus on the nature of the tuning of brain responses to objects along the visual processing hierarchy. Specifically, the project will examine the hypothesis that mid-level features related to generic texture and shape information are tightly linked to high-level properties of objects like their animacy and real-world size. The second aim will characterize the tuning of responses in the ventral stream, by measuring visual and cognitive similarity spaces using both behavior and deep neural networks, and then relating these with neural responses measured with functional neuroimaging. The third aim will introduce a computational framework to relate tuning to topography using Kohonen self-organizing maps. Specifically, the PI will first focus on the large-scale organization of early visual cortex retinotopy as a proof of concept, and then expand this framework to object-responsive cortex. Thus, the PI aims to combine mathematical models of spatial organization with deep neural network representations of visual features, to produce artificial cortical maps that are directly comparable with the cortical organization of the human brain.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.