Machine Learning for Mental State Decoding

This was my dissertation project submitted for the completion of MSc in Advanced Computer Science at the University of Sheffield.

This project involved training predictive models to predict observable mental states from the measured brain activity. Brain activity refers to the blood flow in the neurons inside the brain which is collected as signals during Functional Magnetic Resonance Imaging (fMRI) scans. These scans are performed when participants are asked to perform some actions which stimulates some activity in the brain. This annotation helps in performing supervised learning on neuroimaging data i.e., fMRI.

This project also concentrated on explainability of the deep neural models that were trained during the task of mental state decoding. Certain deep models, that possess the ability to learn complex patterns in the data, often miss out on the explainability aspect which is crucial for making decisions in the health care domain. The project also hints why linear models would leverage the power of explainability with comparable performance in decoding. This trade-off between performance and explainability was the observed and presented during the project.