[Ajoo]

Learning is the ML-equivalent of what statisticians would call fitting a model. It usually involves optimizing a criterium such as the likelihood function of the data, the expected MSE under the posterior (Bayesian) or the margin of a linear classifier (SVM).