SCS Undergraduate Thesis Topics
|Matt Bonakdarpour||Tom Mitchell||Using Machine Learning to Predict Human Brain Activity|
Brain imaging studies are geared towards decoding the way the human brain represents conceptual knowledge. It has been shown that different spatial patterns of neural activation correspond to thinking about different semantic categories of pictures and words. This research is aimed at developing a computational model that predicts functional magnetic resonance imaging (fMRI) neural activation associated with words. The current model has been trained with a combination of data from a text corpus and fMRI data associated with viewing several dozen concrete nouns. Once trained, the model predicts fMRI activation for other nouns in the text corpus with significant accuracy. We hope to explore extensions to the current model in hopes of accurately predicting fMRI activation across subjects and studies. As data is collected with more abstract nouns, new classification techniques will be tested,analyzed and improved. While the current model uses predefined features, this project will explore the usage of various techniques of training and identifying features simultaneously. A survey will also be constructed containing an overview of the machine learning techniques implemented along with the results obtained when used within the model.