Messy coding schemes for prototype selection
John C. Goddard, A.E. Martinez, F.M. Martinez, T. Aljama · 2003
In the present paper, a method is proposed to obtain initial prototypes for nearest neighbour classifiers. The prototypes are obtained by applying an evolutionary program with variable sized chromosomes. Examples are given in the area of speech recognition. The obtained set prototypes can be subsequently used in more general contexts such as a vector quantization codebook or for the centers in a radial basis function network.