Learning Vector Quantization

Synonyms

LVQ

Definition

Learning vector quantization (LVQ) algorithms produce prototype-based classifiers. Given a set of labeled prototype vectors, each input vector is mapped to the closest prototype, and classified according to its label. The basic LVQ learning algorithm works by iteratively moving the closest prototype toward the current input if their labels are the same, and away from the input if not. Some variants of the algorithm have been shown to approximate Bayes optimal decision borders. The algorithm was introduced by Kohonen, and being prototype-based it bears close resemblance to competitive learning and Self-Organizing Maps. The differences are that LVQ is supervised and the prototypes are not ordered (i.e., there is no neighborhood function).