Forming a Multi-Class GLD Classifier

We can form a multiGLD via a generalization of the two-class minimum squared-error procedure.

  • Form augmented data vectors y with higher dimensionality than original data x. [read more]
  • Find a matrix of weight vectors A=[a1...ac] which yields the MSE solution to the conditions:

    (ai)^t • y = 1 for all y in class i.

    (ai)^t • y = 0 for all y not in class i.

     

    Solution?

    A = pinv(Y)B

    where B is a matrix analogous to the margin vector b in the two-class case.

     

 

Back
Contents
Next