Abstract

Artificial ImmuneSystem (AIS)-inspired NeuroEvolution combines the advantages genetic algorithms feature with abstractions of immunological processes.Such processes, applied by immune systems trying to protect organisms from biologically and biochemically hazardous entities, intensely increase the learning performance and accuracy of Multi-Layer Perceptrons performing a stochastic search in a space.This is achieved by applying a combination of immunological operations in each population's evolution cycle, which are clonal selection and somatic hypermutation, negative selection and danger theory.Furthermore, causality plays an important role within the introduced paradigm, as the solution population does not only change from generation to generation, but also within each generation.For the immune system-based operations only the individuals of the current generation do matter.Thus, the in-and outputs a population in consideration processes when learning admittedly have a significance over time for the genetic evolution of a single individual (chromosome).However, all of the immune systembased operations do not need to consider these, as only the current population of genomes matters.Currently, only the already introduced, computationally intelligent Data Mining system "System applying High Order Computational Intelligence in Data Mining" (SHOCID) successfully applies the introduced approach for Artificial Neural Network learning.