If we consider the kind of signal or feedback available to a learning system, then machine learning falls into three categories:
- supervised learning: uses labelled data for training, meaning each piece of data includes both a feature vector x and the desired output y;
- unsupervised learning: uses unlabelled data X for training the model;
- reinforcement learning: the learning system is given a goal and interacts with a dynamic environment.
If instead we consider the desired output of a machine learning system, then we get these categories:
- classification (inputs are assigned to a class, out of a set of predetermined classes)
- regression (inputs are assigned a value from a continuous range)
- clustering (inputs are divided into a series of groups or classes, which are not known beforehand)
- density estimation
- dimensionality reduction.