[editorskit display=”wordcount” before=”Reading Time: ” after=” min”]
Semi-supervised learning is an approach to machine learning problems where only some of the data is labelled (y) for a large amount of input data (x) during training. Thus, semi-supervised learning lies between unsupervised learning (with no labeled training data) and supervised learning (with only labeled training data).
When unlabeled data is used with certain amount of labeled data, there will be considerable amount of improvement in learning accuracy. To label data for learning is expensive and time consuming as it requires skilled domain experts whereas the acquisition of unlabeled data is much inexpensive. Thus, semi-supervised learning has greater practical value as most of the machine learning problems fall into this category. Semi-supervised learning may be done through either transductive learning or inductive learning. The transductive learning infers the correct labels from the given unlabeled data while inductive learning infers correct mapping from x to y.
The supervised learning technique can be used to make best guess and prediction for the unlabeled data, feed that data back into the supervised learning algorithm as training data and use the model to make predictions on new unlabeled data or can use unsupervised learning technique to discover and learn the structure in the input variables. This semi-supervised learning algorithm makes use of at least one of the following assumptions.