David Hardoon
and John Shawe-Taylor
We use Kernel Canonical
Correlation Analysis (KCCA) for detecting brain activity in function MRI by
learning a semantic representation of fMRI brain scans and their associated
time frequency. The semantic space provides a common representation and enables
a comparison between the fMRI and time frequency. We compare the approach
against Canonical Correlation Analysis (CCA) by localising brain regions that
control finger movement and regions that are involved in mental calculation. We
also compare the two approaches on a simulated null data set.
We hypothesis that once a
link can be established between regions of the brain to task one could create a
brain-computer interface were computer related tasks could be activated by
brain "thought" activity.
Spotlight
presentation (PDF file)
Return to Machine Learning and User Interface workshop page.