The following algorithm finds prototypes with high accuracy in all the tested domains [1]. First, we check out lead concepts.

Evolutionary learning

Evolutionary Algorithms are bio-inspired algorithm that uses evolution as a metaphor. I show you an outline below.

EvolutionaryLearning()
initialize-population
    while the termination criterion is not true
     do
         select-parents-for-the-next-generation
         crossover
         mutate

Every generation approximates to the best solution.

Prototype and criticisms

ENPC finds prototypes, but what do prototypes mean? Prototypes are representative data instances from all data or reflect data class. Criticisms are data that is not represented by the prototypes.

Classifier design

Finding prototypes is typically a classifier problem, some solving techniques are K-means clustering, learning vector quantization (LVQ), and Gaussian mixtures.

In fact, this problem is NP-hard, so in order to increase time performance, we use heuristics algorithms but decrease the performance of the solution.

Evolutionary Nearest Neighbour Classifier (ENPC)

ENPC is a heuristic algorithm that finds prototypes, but it doesn’t decrease the performance of the solution. Below is shown an outline.

EvolutionaryLearning()
initialize-population # One random of prototype.
 # number of iterations, accuracy, any combination of previous approaches.
 while the termination criterion is not true
     do
         mutation
         reproduction
         fight
         move
         die

[1] Fernández, F., Isasi, P. Evolutionary Design of Nearest Prototype Classifiers. Journal of Heuristics 10, 431–454 (2004). https://doi.org/10.1023/B:HEUR.0000034715.70386.5b