Unsupervised Learning
A category of learning algorithms based on training [see page 11, without] labelled sample data. This approach tries to automatically find structure in the data by extracting useful features and analyzing its structure.
For example a program can be trained to identify features that differ from sample to sample and cluster similar samples together.
Samples are supplied as: \[ \{{x}^{1}, {x}^{2}, \ldots, {x}^{p} \} \]
The main goal of unsupervised learning is maximising the accuracy rate of the final program. Eg. how accurately can it classify a picture of a cat as a cat instead of a dog.
Unsupervised learning can only do anything useful when there is [see page 1, redundancy] in the input data. Without redundancy it would be impossible to find any patterns of features in the data.