Edinburgh Napier University

  One of the components of an AIS algorithm that most distinguishes it from other paradigms is the use of an affinity funciton. However, for the most part, the implications of choosing a suitable matching rule and the likely effects on the algorithm from an engineering point of view have not been well studied. We develop a simulator that models an idiotypic network that can utilise a number of different shape-space models, which more faithfully model aspects of the real immune system. In each model, an antibody is considered to match any cell which lies within some volume v surrounding the point representation of the antibody (or its complement). The paper’s main contribution is to illustrate how the precise shape of the volume v influences both the dynamics of the resulting networks, and the connectivity of the networks. In particular, the shape of v radically affects the manner in which the entire shape-space can be segmented into tolerant and non-tolerant regions. This has significant bearing for the AIS-practitioner: choosing the correct model may make the difference between being able to find a successful solution to a problem or not.