Pattern Recognition, 4th Edition

15.3. Competitive Learning Algorithms

These algorithms employ a set of representatives w^j, j = 1,…, m.^[3]Their goal is to move each of them to regions of the vector space that are “dense” in vectors of X. The representatives are updated each time a new vector x ∈ X is presented to the algorithm. Algorithms of this type are called pattern mode algorithms. This is a point of differentiation from the hard clustering algorithms discussed in Chapter 14. There the updating of the representatives takes place after the presentation of all the vectors of X to the algorithm. Algorithms of the latter kind are also called batch mode algorithms. It must be emphasized that competitive learning algorithms do not necessarily stem from the optimization of a ...