Deep learning has generated much research and commerc ialization interest recently . In a way, it is the third incarnation of neural networks as pattern classifier s, using insightful algorithms and architectures that act as unsupervised auto-encoders which learn hierarchie s of features in a dataset. After a short review of that work, we will discuss computational approaches for deep learning of behaviors as opposed to just static patterns. Our approach is based on structured non -negative matrix factorizat ions of matrices that encode observation frequencies of behaviors. These te chniques can be used to robustly characterize and exploit diverse behaviors in security applications su ch as covert channel detection and coding. Examples of such applications will be presented.