The Idiotypic Network with Binary Patterns Matching - Artificial Immune Systems - page 107

6 Conclusions

In this paper the new specification of the immune network model is proposed. The
specification differs in a set of fundamental assumptions from the others [2]. Among
the differences we could mention the following: the network is build of the objects
representing types of antibodies and types of antigens instead of just antibodies
or antigens, there is a constant size of repertoire of types of antibodies during the
experiment, strength of stimulation or suppression depends on the number of dif-
ferent types being above the anity threshold and does not depend on their concen-
tration, the concentration is responsible only for the lifetime of the type of antibody
or antigen, relations of stimulation and suppression between types of antibodies as
well as relation between types of antibodies and types of antigens can be controlled
by different anity thresholds. Some of these assumptions are not in accordance
with commonly accepted biological point of view.
The experiments show that proposed rules of dynamics and metadynamics of
the system based on the binary shape-space build a stable network. Three types
of the network behavior can be observed: two of them when the network is not
able to establish itself because all the new objects die immediately after intro-
duction into the system or in the opposite case all the objects once added live
forever. The third type of behavior is the requested one where some of the anti-
bodies live longer but a recruitment of the new ones is also performed and this
way the stable network is build. It was observed that the chances for stable net-
work strongly depends on the type of anity measure. For some of the measures
it was impossible to tune the anity thresholds successfully. A new transforma-
tion T operator was proposed which significantly influenced the properties of the
measures and when applied gave a set of three new measures resulting in more
regular and predictable behavior of the network. Further work could be focused
on testing all the selected ecient measures applied in the network where a set
ofnewpatternsispresented.

References

1. V. Batagelj and M. Bren. Comparing resemblance measures. Journal of Classification ,
12(1):7-90, 1995.
2. H. Bersini. Revisiting idiotypic immune networks. In W. Banzhaf, T. Christaller,
P. Dittrich, J. T. Kim, and J. Ziegler, editors, Advances in Artificial Life, 7th European
Conference, ECAL 2003, Dortmund, Germany, September 14-17, 2003, Proceedings ,
volume 2801 of LNCS , pages 164-174. Springer, 2003.
3. L. N. de Castro and J. Timmis. Artificial Immune Systems: A New Computational
Approach . Springer-Verlag, London. UK., 2002.
4. J. C. Galeano, A. Veloza-Suan, and F. A. Gonzalez. A comparative analysis of artificial
immune network models. In H.-G. Beyer and U.-M. O'Reilly, editors, Genetic and
Evolutionary Computation Conference, GECCO 2005, Proceedings, Washington DC,
USA, June 25-29, 2005 , pages 361-368. ACM, 2005.
5. J. E. Hunt and D. E. Cooke. Learning using an artficial immune system. Journal of
Network and Computer Applications , 19:189-212, 1996.