Decoding brains by paying attention: An attention-based fMRI task state decoding deep network architecture

One of the core goals in the field of cognitive neuroscience is to decode task state fMRI data. Task decoding is the process of taking neuroimaging data and determining the task that was performed when that data was collected. A large volume of work used for task decoding is done in pursuit of creating a deep learning model for task prediction. Typically these models will include either handcrafted features or data driven approaches for downscaling the input features in successive layers. In this thesis, we explore and compare the effectiveness of linear, graph-based and attention-based methods for hierarchical classification. Furthermore, we propose a new attention-based network architecture which showcases superior performance to all of our baseline architectures without the use of handcrafted features on several neuroimaging datasets.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• Roxas, Francis

Contributors dc:contributor

• Koyejo, Oluwasanmi

• Transformers, Graph Neural Networks, Graph Attention Networks, Attention, Machine Learning, Deep Learning, Brain Decoding

Statement dc:rights

• Copyright 2021 Francis Roxas

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/113077

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/113077