Meta-stable Memory in an Artificial Immune Network

This paper describes an artificial immune system algorithm which implements a fairly close analogue of the memory mechanism proposed by Jerne [1] (usually known as the Immune Network Theory). The algorithm demonstrates the ability of these types of network to produce meta-stable structures representing populated regions of the antigen space. The networks produced retain their structure indefinitely and capture inherent structure within the sets of antigens used to train them. Results from running the algorithm on a variety of data sets are presented and shown to be stable over long time periods and wide ranges of parameters. The potential of the algorithm as a tool for multivariate data analysis is also explored.