The simplest definition of unsupervised learning is that it is the process of providing some sort of structure to unstructured data by discovering and utilizing the inherent patterns of the data. If data is not produced by some random process, it will have some patterns between its data elements in its multidimensional problem space. Unsupervised learning algorithms work by discovering these patterns and using them to provide some structure to the dataset. This concept is shown in the following diagram:

Note that unsupervised learning adds structure by discovering new features from the existing patterns.