Georg Martius (Project Leader),
Eric Lacosse
In this project, we look at the prediction of brain activity, measured by activation maps in fMRI, that occur while performing a task from the activation that occurs during resting. The latter means the patients are asked to rest in the fMRI scanner. Prior work has claimed significant prediction capabilities.
In the project, we check existing methods and find that with a stringent machine learning methodology many of the results are actually not better than a very simple baseline and did indeed not predict anything meaningful. We find that with improved techniques some of the prediction performance can be regained.