Frank Noé - Freie Universität Berlin Markov state models (MSMs) have become a widely-used technique to analyze high-throughput molecular dynamics (MD) data. I will discuss generalizations of MSMs and deep learning models of kinetics. I will illustrate how these techniques can be used to understand the essential character of massive MD data, perform efficient dimension reduction in dynamical systems, and how they can be exploited to drive simulations to sample rare events.