An Investigation of a Methodology for the Development of Artificial Immune Systems: A Case-Study in Immune Receptor Degeneracy

For nearly half a century, biology has provided a rich source of inspiration for computing systems, the result of which is a large body of work that includes biologically-inspired algorithms such as artificial immune system (AIS). Despite the effort that has been invested in engineering these algorithms, relatively little research has focused on how best to extract the underlying biological properties from which inspiration is taken. This thesis aims to address this failing by following a more principled approach to the development of an AIS. We begin by exploring the current state of AIS, and examine how immunology has been used to inspire AIS to date. This leads us to identify a methodology for developing AIS that incorporates a number of explicit modelling stages to extract the key features of the biological system. An examination of the immunological literature identifies our immune inspiration: immune receptor degeneracy and the mechanism of patterns of response to provide immune specificity. Our first step in developing an AIS based on these properties is to build an agent-based simulation to explore them free of any engineering application bias. We then investigate the idea of tunable activation thresholds for immune detectors, which results in the generation of a single pattern of response from a detector population when subject to a stimulus. Using the insight gained from these investigations, we construct a framework to allow patterns of degenerate tunable detectors to be incorporated into AIS as a data pre-processing stage. This framework it then instantiated for a simple pattern classification AIS. To conclude we analyse the process we have followed to develop our AIS and assess the benefits and drawbacks of the approach we have taken, showing how a more principled approach can be applied to the design of biologically-inspired algorithms.