Training a model to maximize the probability it assigns to observed data, a standard approach in supervised learning.