Our group develops statistical machine learning algorithms to analyze fMRI data. We are specifically interested in algorithms that can learn to identify and track the cognitive processes that give rise to observed fMRI data.
Watch the video demonstration of our computer program decoding which candidate word a person is thinking about, based only on the neural activity captured in their fMRI data. The program was trained using fMRI data from other people, indicating that our different brains encode word meanings in quite similar ways.
What the data look like: In one fMRI study we trained our algorithms to decode whether the words being read by a human subject were about tools, buildings, food, or several other semantic categories. The trained classifier is 90% accurate, for example, discriminating whether the subject is reading words about tools or buildings.
The following figure shows, for each of three different subjects, the degree to which different brain locations can help predict the word's semantic category. Red and yellow voxels are most predictive. Note the most predictive regions in different subjects are in similar locations.
Acknowledgements: We thank the W.M.Keck Foundation and the U.S. National Science Foundation for their generous support of this research.