Machine learning is based on several fundamental types:
When you manually "teach" system to recognise aome objects. Trained data is olways labeled so machine does predictions based on previous experience with previous labels.
Training data goes to model in order to consume Testing data later on and provide Output
When you provide Data to the nodel and it should recognize patterns on its own. It devides data into clusters by connections among data.
For example, Facebook friend recommendations does the same thing. It maps your friend list and recommend you to follow those who are into your data cluster.
Machine evaluate every move in order to increase its chances to win (in terms of some game)
You can "Teach" your model with Caffe, torch, k, e.t.c
Core ML allows you to load a Pre-Trained Model and easily convert it for work in Xcode (.ml file)
Core ML allow you to do predictions