I started my thesis by improving decomposition techniques for fMRI. We introduced new scalable algorithms to extract functional networks from fMRI recordings. These algorithms can decompose recent massive datasets (eg HCP) in a reasonable time (~10h). They have a theoretical grounding and can be applied to other problems (e.g. collaborative filtering) as well.
I am now involved in setting up richer models for these datasets, still keeping scalability in mind.
Please visit amensch.fr for more details on my work !