Results of experiments disscussed in paper
"Adaptive clusters formation in an Artificial Immune System"

S. T. Wierzchoń, U. Kuzelewska submitted to National Conference on Evolutionary Computation and Global Optimization, 2002.
Page includes graphs of experiments refered in the paper.

Back to Main Page

Three sets of training data were examined. Results are presented on figures 1 - 3. The immune networks produced by our algorithm are:

  1. stable in time (their size doesn't exceeds the size of the training set)
  2. they adaptively fit the value of the network affinity threshold, or NAT derived by Timmis (consult [1])
  3. the algorithm does not require control parameters

Experiment #1

Training data Final immune network (animated version)
Number of correctly recognized cells (solid line) and the network size (dashed line) evolution NAT scalar evolution

Experiment #2

Training data Final immune network
Number of correctly recognized cells (solid line) and the network size (dashed line) evolution NAT scalar evolution

Experiment #3

Training data Final immune network
Number of correctly recognized cells (solid line) and the network size (dashed line) evolution NAT scalar evolution

Made with CSS Copyright © 2002 AIS at ICS PAS, Warsaw, Poland. All rights reserved.
Information about this page format is available HERE.