Yaoda Xu - US grants

Many of our behaviors depend on the ability to rapidly recognize objects in the real world. Yet, as effortless as visual object perception seems to be, even for young infants rapidly learning the names of surrounding objects, this capability eludes the most sophisticated computers and devices. In fact, many details of this process remain unknown despite decades of research progress in neuroscience and cognitive psychology. Understanding how the human brain, which is a physical device that performs computations, recognizes objects is therefore a useful and important endeavor. Recent advances in brain imaging technology, especially functional magnetic resonance imaging (fMRI), have now made it possible to safely examine the brain mechanisms in everyday adult human observers. One basic question concerns how neurons represent complex visual objects that typically consist of distinct parts arrayed in a particular configuration. For example, a bicycle has wheels, a frame, and handlebars arranged in a certain way that enables people, such as car drivers, to quickly recognize one on the road. With support from the National Science Foundation, Dr. Yaoda Xu and Dr. Marvin Chun are using fMRI to probe detailed brain activity while observers perform visual recognition tasks in the MR scanner. In particular, this project focuses on how specific object parts and part configurations are represented and distinguished from others. This knowledge will advance our understanding of how the human brain recognizes visual objects.

The work should have significant implications for theories of visual object perception and especially for understanding the impact of brain damage on visual recognition abilities (agnosias). Furthermore, an understanding of how the brain recognizes objects should facilitate the development of sophisticated computer systems that can recognize and learn visual objects in our environment. During the course of this project, student collaborators will gain training in advanced functional brain imaging technologies and experimental methods to study human behavior.

In everyday life, perceivers are confronted with continuous and overwhelming influxes of visual information from the environment. To extract the most relevant visual information to guide behavior and thought, a visual system is faced with two challenges. One challenge is to select discrete units of visual information from competing inputs from the environment (e.g., detecting an approaching vehicle on the road). A second challenge is to integrate visual information initially processed in separate visual areas (such as color, shape and motion) to achieve a single coherent visual percept (e.g., a red moving car on the left). Although vision research has primarily focused on feature processing in visual cortex, brain lesion and brain imaging studies indicate that the parietal cortex plays an essential, but at present largely mysterious role in visual information selection and integration.

With support of the National Science Foundation, Drs. Yaoda Xu and Marvin Chun at Yale University will use a brain imaging technique, functional magnetic resonance imaging, to study visual processing in the parietal cortex of healthy human observers. This work will provide with a broad understanding of the role of the parietal cortex in visual perception. Because parietal cortex is often damaged by stroke, causing deficits in visual information selection and integration, the outcome of this research may provide clues to aid patient rehabilitation after parietal lesions. The research may also provide significant insights to computer vision and the design of sophisticated artificial visual systems. Finally, the project will provide research experience for both graduate and undergraduate students.

DESCRIPTION (provided by applicant): Our visual world is filled with clusters (or ensembles) of objects. For many visual tasks, such as attending and selecting specific objects for thoughts or actions, we need to individuate and encode distinctive features from single objects. Such object-specific processing has been the core of cognitive and neuroscientific research for the past few decades. In contrast, there are also ample occasions when our visual system extracts summary statistics from an object ensemble without representing each object in the ensemble in great detail. Such object-ensemble representation can help us rapidly segment an otherwise crowded visual scene and allow us to zoom in on specific objects of interest. Object-ensemble representation thus complements and guides object- specific processing and allows our visual system to overcome the capacity limitation of object-specific processing. Yet despite the common occurrence of object ensembles in our visual world, the cognitive and neural mechanisms underlying object-ensemble processing remains poorly understood. Using fMRI, this proposal will first identify where in the human brain object ensembles are represented, under what attentional condition this representation can be formed, and then examine the nature of the neural object ensemble representation by studying how mean and variance of ensemble features are represented. The proposed studies will be one of the first to systematically investigate how real world object ensembles are processed in the human brain. Besides furthering our understanding of the connection between brain and behavior, results of these studies will have significant impact on current cognitive and neural theories on visual object perception. PUBLIC HEALTH RELEVANCE: The proposed research will help us understand the various constrains and limitations in normal visual cognition. It also has significant implications for understanding the impact of brain damage on visual object perception and recognition.

PROJECT SUMMARY Central to human cognition is the ability to interact adaptively with our visual environment by collecting and accumulating pertinent information to guide thoughts and behavior. Recent work has pinpointed the significant role of posterior parietal cortex (PPC) in supporting adaptive visual processing. Building upon two recent reviews on PPC and two sets of exciting preliminary findings on visual representation in human PPC involving attention and visual working memory (VWM), the present proposal aims to document the nature of visual processing in PPC with the same rigor and precision as those used to study visual representation in OTC. The overarching hypothesis here is that task-related factors play a more prominent role in shaping visual representation in PPC than OTC and the key to understand the intricate details of visual representation in PPC is through its interactions with task-related factors. Using fMRI pattern decoding and representational similarity analyses to document PPC representational structure in humans, the present proposal examines two fundamental aspects of adaptive visual processing: The contributions of attention and task to adaptive visual representation in PPC in Aim 1, and the nature of VWM representation in PPC in Aim 2. To understand the neural underpinnings of PPC visual processing, using voxel overlap analysis, in all the studies proposed, we will examine whether the same or distinct PPC neuronal populations contribute to different aspect of PPC visual processing. The proposed studies will provide for the first time an in-depth and systematic understanding of the nature of PPC visual representation in the human brain and its interaction with important task-related factors. They shall provide foundational knowledge regarding how adaptive visual processing is accomplished in the human brain.