EEG source localization is hard. Current state of art is beamforming. We took a simultaneous EEG-fMRI dataset, used fMRI to find true source and measured how well a beamformer can predict the Broadmann area where the activity is. Turned out that it could not (random prediction level). Gagandeep then developed an approach based on autoencoders that achieved higher-than-random classification ability.