LVQ is another kind of competitive network that uses competitive and linear layer together (see past posting for competitive network). The winning neuron of the first stage is become a subclass. Then the second layer combines it into a single class. Many researcher use this kind network for signature identification.

The learning of LVQ is supervised learning. It needs a target as a basis of learning process. Use this script (LEARNING VECTOR QUANTIZATION) to practice your matlab skill in creating a LVQ network. Try to understand the syntax of lvq function by typing “help newlvq” on command window.

The above picture shows matlab respons to our train script function. You can see the IW and LW part of LVQ network using this script.

>> samplelvq.iw{1,1}

ans =

-0.1259 -0.0280

0.0862 0.1584

-0.0351 0.0306

>> samplelvq.lw{2,1}

ans =

1 1 0

0 0 1