Theoretical foundations of artificial immune systems

Artificial immune systems (AIS) are a special class of biologically inspired algorithms, which are based on the immune system of vertebrates. The field constitutes a relatively new and emerging area of research in Computational Intelligence that has achieved various promising results in different areas of application, e.g., learning, classification, anomaly detection, and (function) optimization. An increasing and often stated problem of the field is the lack of a theoretical basis for AIS as most work so far only concentrated on the direct application of immune principles. In this thesis, we concentrate on optimization applications of AIS. It can easily be recognized that with respect to this application area, the work done previously mainly covers convergence analysis. To the best of our knowledge this thesis constitutes the first rigorous runtime analyses of immune-inspired operators and thus adds substantially to the demanded theoretical foundation of AIS. We consider two very common aspects of AIS. On one hand, we provide a theoretical analysis for different hypermutation operators frequently employed in AIS. On the other hand, we examine a popular diversity mechanism named aging. We compare our findings with corresponding results from the analysis of other nature-inspired randomized search heuristics, in particular evolutionary algorithms. Moreover, we focus on the practical implications of our theoretical results in order to bridge the gap between theory and practice. Therefore, we derive guidelines for parameter settings and point out typical situations where certain concepts seem promising. These analyses contribute to the understanding of how AIS actually work and in which applications they excel other randomized search heuristics