In feature discovery, the focus is less on the output of an algorithm than on the internal structures that an algorithm learns and what these factors reveal about the input data. For example, an algorithm exposed to a variety of different handwriting samples could be used for classification. At the same time, the internal structure learned by the algorithm may yield interesting insights into the features that make up handwriting, and such insights are the domain of feature discovery.
Care should be taken not to confuse feature discovery with feature selection, which is a synonym for dimensionality reduction.
- alias