A new perspective to cognitive neuroscience: understanding the information processing strategies of human brain

Duan, Yaocong (2024) A new perspective to cognitive neuroscience: understanding the information processing strategies of human brain. PhD thesis, University of Glasgow.

Abstract

Visual categorization is one of the most fundamental and crucial cognitive functions of the human brain. An unsolved mystery about human visual categorization ability is its remarkable efficiency and generalization. These abilities rely on the brain’s sophisticated system that actively extracts and processes the task-relevant features, through an integration of both bottom-up processing of visual input and top-down information of task context. This thesis argues that the complexity of features represented in the brain has not been adequately considered. Reconstructing how the brain actively transforms complex visual input into taskrelevant features could shed light on the underpinnings of visual efficiency and generalization.

In this thesis, I employed a novel approach to tackle this problem. By precisely controlling the visibility of individual pixels in images and utilizing high spatial and temporal resolution neuroimaging data, combined with information-theoretic analysis framework, I reconstructed the specific pixels that are collectively represented by neural activity, thereby dynamically revealing the visual content (i.e. features) within images that are represented by brain and which brain regions, at what time, transform the complex visual input into taskrelevant features.

Moreover, I also introduced a new information-theoretic measure, termed Samplewise Mutual Information (SMI), which quantifies the single-trial relationships between two variables. By applying SMI to characterize the single-trial relationship between participant behavior and the image samples, and employing Non-negative Matrix Factorization (NMF) clustering algorithm to learn the local parts of pixels that collectively influence the participant’s behavior, I identified the finer components of the features that can better predict the participant’s behavior. These feature components could serve as the minimal processing units by the brain.

These works advance our understanding of information processing strategies of the human brain for visual categorization tasks and open up a new analysis framework for future research.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Additional Information: Supported by funding from the China Scholarship Council.
• Subjects: R Medicine > RC Internal medicine > RC0321 Neuroscience. Biological psychiatry. Neuropsychiatry; T Technology > TA Engineering (General). Civil engineering (General)
• Colleges/Schools: College of Science and Engineering > School of Engineering
• Funder's Name: China Scholarship Council (CSC)
• Supervisor's Name: Schyns, Professor Philippe and Ince, Dr. Robin
• Date of Award: 2024
• Depositing User: Theses Team
• Unique ID: glathesis:2024-84568
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 18 Sep 2024 09:30
• Last Modified: 18 Sep 2024 09:33
• Thesis DOI: 10.5525/gla.thesis.84568
• URI: https://theses.gla.ac.uk/id/eprint/84568