© University of Kent - Contact | Feedback | Legal | FOI | Cookies
Hierarchy and convergance of immune networks: Basic ideas and premilinary results
L.N de Castro and J Timmis
In J. Timmis and P.J. Bentley, editors, 1st International Conference on Artificial Immune Systems, pages 182-196, University of Kent at Canterbury, September 2002. Unversity of Kent at Canterbury Printing Unit.Abstract
aiNet is an artificial immune network model originally developed to perform automatic data compression. Combined with graph theoretical and statistical clustering techniques, aiNet is a powerful data clustering and classification tool. However, the original aiNet model suffers from the lack of a well-defined stopping criterion and an ad hoc approach to parameter initialization, prior to the training process. This paper has two main goals. First, by assessing convergence criteria employed in a class of artificial neural networks, a suitable stopping criterion can be created for aiNet. Secondly, the paper demonstrates that through the use of a cooling schedule for some of these user-defined parameters, it is not only possible to reduce the importance of their initial values, but also this leads to possible derivation of a hierarchical tree of immune networks. Due to the very limited space available, only the basic ideas of a novel convergence criterion, and an approach to develop a tree of aiNets will be presented, together with an illustrative example.
Bibtex Record
@inproceedings{1506,
author = {de Castro, L.N and Timmis, J},
title = {Hierarchy and Convergance of Immune Networks: Basic Ideas and Premilinary Results},
month = {September},
year = {2002},
pages = {182-196},
keywords = {determinacy analysis, Craig interpolants},
note = {},
doi = {},
url = {http://www.cs.kent.ac.uk/pubs/2002/1506},
publication_type = {inproceedings},
submission_id = {21900_1031842170},
ISBN = {1902671325},
booktitle = {1st International Conference on Artificial Immune Systems},
editor = {Timmis, J. and Bentley, P.J.},
address = {University of Kent at Canterbury},
publisher = {Unversity of Kent at Canterbury Printing Unit},
refereed = {yes},
}