In Learning theory, a hyperparameter refers to a parameter in the learning algorithm itself, which isn't learned by the algorithm, but is chosen before running it. Hyperparameters may be learned (or hyperlearned), via methods of Model selection.