Analysis/Probability Learning Seminar

We continue talking about Ramon van Handel's new tutorial on Structured Random Matrices. This week I will present the proof of Van Handel's nearly sharp improvement of Latala's inequality. Latala's inequality yields a dimension-free bound on the spectral norm of Gaussian matrices with independent entries (but not necessarily i.i.d., the entries can have arbitrary variance pattern). Van Handel's stronger version of the inequality is also optimal for many structured matrices. Speaker(s): Liza Rebrova (UM)