In this talk, we will first review log-log blow up dynamic of mass critical NLS, and then discuss how to do a rough but structured (random) perturbation via probabilistic methods. Based on Joint work with Dana Mendelson (University of Chicago).