Delineating the neural computational network for object recognition

Project Summary/Abstract A core function of the human brain is object recognition, which underlies many higher-order functions like memory and emotion. Deficits in recognizing objects are associated with various neurological disorders, such as visual agnosia, schizophrenia, and Alzheimer’s disease. However, the neural mechanisms for object recognition remain

elusive. Specifically, little is known about how the brain translates visual inputs of objects into meaningful semantics. Previous studies have proposed a hierarchical neural network for object processing, critically including the visual temporal cortex (VTC) and the downstream medial temporal lobe (MTL). While recent single-

neuron studies in non-human primates evidenced an axis-based visual feature encoding in the VTC, human MTL neurons have long been characterized to carry a sparse and selective code for individual exemplars (exemplar- based coding). Yet the process by which visual feature representations in the VTC are transformed into semantic

representations of abstract labels in the MTL remains unknown. This project aims to address this question by examining the neural computations and dynamics within the VTC- MTL neural network during object recognition. We will utilize intracranial recordings at both individual neuron and neural-circuit levels across different brain areas, coupled with sophisticated computational algorithms. Three

distinct neural coding models will be surveyed across the VTC and MTL, including the axis-based feature model, the exemplar-based model, and the region-based feature model (a novel model proposed in our recent studies; K99 AIM 1). VTC-MTL interactions and dynamics that are critical to object recognition will be identified (K99 and

R00 AIM 2). The derived results and analysis pipeline from the K99 phase will then be utilized to investigate how different neural models transition from one to another along the VTC-MTL pathway to achieve the representation transformation (R00 phase). The central hypothesis is that different coding models are employed at different

stages of object processing, and the novel region-based feature coding serves as an intermediate step that bridges the axis-based coding in the VTC and the exemplar-based coding in the MTL. This proposal will be conducted at Washington University in St. Louis (WUSTL), a top-rank research university that offers excellent scientific support and career training. A team of established mentors will provide necessary

training in: iEEG data recording and analysis (Drs. Brunner and Willie), inter-areal interaction analyses (Dr. Rutishauser), neural networks and cognitive neuroscience more broadly (Dr. Hershey), computational modeling and statistics (Dr. Wang), and career development (all mentors). While the primary aims of this proposal is to

provide new insights into object recognition in humans across multiple scales, the anticipated outcomes have the potential to inspire new therapeutic interventions for disorders involving impaired object recognition. This K99 award will critically facilitate the success of the proposed research and the applicant’s transition to independence.

1