The settings chosen before training that govern how a model learns, as distinct from the parameters it learns.
