Indepdent components analysis (ICA) and other factor analytic tequniques are useful for investigating brain networks in resting state fMRI. ICA, in particular, has become a standard component in the toolbox for neuroimaging scientists. Group ICA is simply the application of ICA to a collection of fMRI resting state scans. A group ICA model typically requires two dimension reduction steps: one mathematically required and one done for computational convenience. In addition, standard group ICA does not scale to large modern data sets, such as those in the ADHD 200, 1K Functional Connectome and ABIDE data sets. Recent work by Ani Eloyan and other SMART group members gives a scalable, likelihood based version of ICA that eliminates the unncessary SVD. To the left, we show a proof of concept image yielding networks (notor and default mode) estimate from 100 resting state scans from the 1K Functional Connectome data set. These represent a comprimise-free use of ICA in resting state that removes unnecessary data reduction. Further work is extending the algorithm to take advantage of parallel processing. A current goal of the group is to use large public data sets to create a definitive functional brain atlas.