The standard assumption in classification is that the training data are independent and identically distributed. Indeed, this assumption is so pervasive that it is often applied without mention. In this dissertation, we ...