This is joint work with Kiran Vodrahalli and Lydia Liu, under the guidance of Barbara Engelhardt.

We introduce a new lens through which to analyse time-series brain data, with emphasis on sparse, low-dimensional joint representations of MEG and EEG data. Our main contribution is empirical validation suggesting that multimodal sparse CCA is able to achieve a low-dimensional representation of time series brain data which retains predictive and generative power.

We also validate our methods using EEG representations of fMRI data: by using spatial information encoded by paired EEG-fMRI data with sparse CCA, we verify that these joint representations have predictive power by training SVMs to distinguish states of attention from base states.

Check out our project page here for more information.