Semi-supervised learning
In supervised learning, the training data typically have INPUT \ (X ^ R & lt \) , there Output \ (\ R & lt Hat {Y} ^ \) .
However, in practice, data collection is not difficult, but doing data tag takes a lot of effort.
So semi-supervised learning, is to use a lot of data unmarked \ (the X-^ U \) , generally U >> R.
transductive learning: unlabeled data is the Data Testing,
Inductive Learning: These unlabeled data do not do testing data.
Semi-supervised learning, although unlabeled data does not provide direct training samples, but its input distribution can help to better classify demarcation.
But the semi-supervised learning using unlabel way is often accompanied by a number of assumptions, in fact, semi-supervised learning there is no use, depending on your assumption is not realistic character / fine inaccurate.