Learning vector quantization
Learning vector quantization, English learning vector quantization ( short LVQ ) is a method from the field of artificial neural networks. It is used for classification of non-linear problems.
VQ = vector quantization: A method for unsupervised clustering
LVQ = learners VQ
Classification: Picture in K classes training set P
P = {( xi, yi) in 1 x { K} |. } I LVQ network: given by prototypes (w1, Y1) ... (wk, Yk ) in { 1. } K defined mapping x → Yi with | x- wi | minimal, so after winning the WTA ( Winner Takes It All )
LVQ1: there are created and their weight vectors randomly initialized with random patterns of each class, by the priorities of the relevant Class or otherwise useful for classes 1 to k one or more neurons. Then, just as with vector quantization, the pattern presented and the calculated one winner neuron with the smallest Euclidean distance to the input. The learning rate [0,1] is either constant or decreasing during the proceedings to enforce convergence.
Algorithm:
Init wj repeat Choose (xi, yi) determine winners (wk, yk ) wk = wk ( xi -wk ) if yi = Yk ( yi if the class represented by wk ) wk (xi -wk ) otherwise The convergence of LVQ1 is not proven; There are problems in practice with overlapping data.
- Neuro computer science
- Machine Learning