This completes the overview of three of the most commonly used unsupervised learning techniques:
K-means for clustering fully observed features of a model with reasonable dimensions
Expectation-maximization for clustering a combination of observed and latent features
Principal components analysis to transform and extract the most critical features in terms of variance for linear models
Manifold learning for non-linear models is a technically challenging field with great potential in terms of dynamic object recognition [4:18].
The key point to remember is that unsupervised learning techniques are used:
By themselves to extract structures and associations from unlabelled observations
As a preprocessing stage to supervised learning in reducing the number of features prior to the training phase
The distinction between unsupervised and supervised learning is not as strict as you may think. For instance, the K-means algorithm can be enhanced to support classification.
In the next chapter, we will...