The classical define is that it is the field of algorithms that infer the problem they compute from test text . An examples of this is the " Theo of the Learning " journal which is arguably the seminal computer learning paper . In this frame it is meant to be program Science ' s answering to Buddhism.
This it is maybe very narrow here , as the order of inferencing a product from examples makes the most work with Classification or regression methods and making less use with bunch or other questions .