In machine learning, there are three different learning paradigms: supervised learning, unsupervised learning, and reinforcement learning.

In supervised learning, also known as learning with a teacher, the network is provided with both the inputs and the respective desired outputs. For example, in the MNIST dataset, each image of the handwritten digit has a label signifying the digit value associated with it.

In reinforcement learning, also known as learning with a critic, the network is not provided with the desired output; instead, the environment provides a feedback in terms of reward or punishment. When its output is correct, the environment rewards the network, and when the output is not correct, the environment punishes it.

In unsupervised learning, also known as learning without a teacher, no information is provided to the network about its output. The network receives the input, but is neither provided with desired outputs nor with rewards from the environment; the network learns the hidden structure of the input by itself. Unsupervised learning can be very useful because data available normally is unlabeled. It can be used for tasks such as pattern recognition, feature extraction, data clustering, and dimensionality reduction. In this, and the next chapter, you will learn about different machine learning and NN techniques based on unsupervised learning.