2. Clustering (or partitioning)

2.1 Clustering applications

The aim of clustering is to group a series of records, be they vectors, images or more complex objects, into groups, or clusters. Generally speaking, the aim of this type of algorithm is to ensure that objects assigned to the same cluster are similar to one another. It's simply a matter of dividing the objects into groups that are as homogeneous as possible. However, we must be aware that this definition is fragile, and even insufficient:

• clustering involves choosing a similarity criterion between objects. When considering real, low-dimensional vectors, this choice will be relatively easy: Euclidean distance, for example. If, on the other hand, clustering is to be carried out on high-dimensional objects (images,...