Expanding the Capabilities of Voxelwise Modeling for Naturalistic Brain Decoding

Ryan Ong

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2023-5
January 15, 2023

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2023/EECS-2023-5.pdf

Autonomous navigation systems have yet to reach the advanced capabilities of the human brain. Further understanding of the brain, through neuroscience, has the potential to facilitate the improvement of modern artificial intelligence systems. Neuroscience is an emergent field that takes a principled approach for investigating the brain. This thesis weaves three projects together into a narrative that follows the pipeline of the voxelwise modeling framework. Voxelwise modeling is a modern, data science-inspired approach for fMRI brain encoding (and, consequently, decoding in the reverse direction) within naturalistic environments. We detail our efforts to customize a driving simulator, CARLA (Unreal Engine 4), for brain decoding/encoding stimuli. Next, we propose and verify a pipeline for non-linearly transforming stimuli into semantic features. Finally, we explore fitting voxelwise encoding models, with multiple feature spaces, to find cortical representations of timescale selectivity.

Advisor: Allen Yang


BibTeX citation:

@mastersthesis{Ong:EECS-2023-5,
    Author = {Ong, Ryan},
    Title = {Expanding the Capabilities of Voxelwise Modeling for Naturalistic Brain Decoding},
    School = {EECS Department, University of California, Berkeley},
    Year = {2023},
    Month = {Jan},
    URL = {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2023/EECS-2023-5.html},
    Number = {UCB/EECS-2023-5},
    Abstract = {Autonomous navigation systems have yet to reach the advanced capabilities of the human brain. Further understanding of the brain, through neuroscience, has the potential to facilitate the improvement of modern artificial intelligence systems. Neuroscience is an emergent field that takes a principled approach for investigating the brain. This thesis weaves three projects together into a narrative that follows the pipeline of the voxelwise modeling framework. Voxelwise modeling is a modern, data science-inspired approach for fMRI brain encoding (and, consequently, decoding in the reverse direction) within naturalistic environments. We detail our efforts to customize a driving simulator, CARLA (Unreal Engine 4), for brain decoding/encoding stimuli. Next, we propose and verify a pipeline for non-linearly transforming stimuli into semantic features. Finally, we explore fitting voxelwise encoding models, with multiple feature spaces, to find cortical representations of timescale selectivity.}
}

EndNote citation:

%0 Thesis
%A Ong, Ryan
%T Expanding the Capabilities of Voxelwise Modeling for Naturalistic Brain Decoding
%I EECS Department, University of California, Berkeley
%D 2023
%8 January 15
%@ UCB/EECS-2023-5
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2023/EECS-2023-5.html
%F Ong:EECS-2023-5