Ability to identify, among a set of input data, instances of objects belonging to a certain class.
In order to identify the objects (patterns), a model of the class is needed. This model is built on a prior knowledge, usually given by a training data set. In this case the prior knowledge is implicit and will be coded in some way by the system aimed to identify the patterns.
For instance, a neural network will modify itself depending on the training, thus coding the information given by the data in its own structure.