猕猴物体识别的视觉神经机制

One pivotal function of the brain is to identify objects in the surroundings based on sensory inputs. Primates heavily rely on their visual system to extract object-related information. Inferotemporal (IT) cortex is believed to be the brain region responsible for object recognition. In the past decades, we have gained some basic understandings on the rules governing the code for objects in IT cortex. However, several important questions remained unanswered, for example: is information related to individual objects represented by individual IT neurons or neuronal populations? As a large brain region, could IT be further divided into multiple smaller regions? The existence of such controversies is largely due to the following reason: the concept of object is rather abstract and we don’t have very good tools to quantitatively describe and generate object images. We will use new tools recently developed in the field of artificial intelligence, such as object recognition network and generative network, to describe and generate object images in a quantitative way. With that help, we will perform functional MRI and electrophysiology experiments in awake monkeys to systematically explore the neural code for objects in IT. The proposed research will not only shed light on the neural basis of object vision, but it will also facilitate the development of brain-inspired intelligence.

大脑的重要功能之一是通过感觉输入识别外界的物体。灵长类（包括我们人类）十分依赖于视觉系统来提取物体相关信息。负责物体识别的视觉脑区是下颞叶皮层。过去几十年的研究使我们对下颞叶神经元编码物体的方式有了基本的了解，然而，仍然有很多基本问题尚待回答。比如：在下颞叶，单个物体的信息是由极少量细胞编码还是有大量神经元群体编码？下颞叶皮层作为一个较大的脑区是否可以按照功能细分为多个小区？研究这一问题的困难之处在于物体的概念比较抽象，我们之前没有很好的手段定量地刻画和生成物体图像。申请人将借助近几年人工智能领域发展出的新工具：如物体识别网络、生成性网络等，对物体进行定量的刻画和生成，并通过清醒猴的功能核磁共振和电生理实验系统研究下颞叶神经元对物体的编码方式。该研究成果不仅有助于揭示大脑对外界视觉信息编码及物体识别的普遍原理，而且对研发新型类脑人工智能的算法及模型有潜在的促进作用。

世界由形形色色的物体构成。在和外部世界打交道的过程中，高等动物进化出复杂而精细的视觉系统从环境中提取物体信息。灵长类负责物体识别的脑区是下颞叶皮层。由于物体属性是很抽象的概念，如何研究这一脑区的编码性质始终是一个难题。本项目使用功能核磁共振结合电生理记录研究下颞叶如何加工处理物体信息。本项目的主要目标是建立灵长类颞叶物体信息的表征地图。通过深度学习的方式提取物体空间的主要特征，并通过功能核磁共振对这些特征的分布方式进行定量刻画，对3只猕猴和4名人类被试建立了颞叶物体特征表征的全景地图，并结合电生理实验对前人研究未涉及的脑区进行了功能刻画。综上，本课题产生了一系列原创性成果，从宏观和微观角度对物体信息的表征和分布模式提供了全新的素材和视角，为新一代类脑人工智能系统提供了数据上的支持和思路上的启发。