Parallelizing an Immune-inspired Algorithm for Efficient Pattern Recognition

In recent years, there has been a growth of investigation into the use of the mammalian immune system as a source of inspiration and metaphor for computational tasks. One avenue of this investigation has been the exploration of the learning capabilities demonstrated by these biological systems. A second appealing aspect of biological immune systems is their inherent distributedness. This paper examines these two strains by proposing modifications to an existing immune-inspired algorithm to provide a more efficient, parallel version. This investigation is offered by way of a case-study in the gains paralleization can provide to current Artificial Immune Systems and as a proofof-concept in the field of parallel immune learning. INTRODUCTION In recent years, there has been a growth of investigation into the use of the mammalian immune system as a source of inspiration and metaphor for computational tasks (deCastro and Timmis 2002). One avenue of this investgation has been the exploration of the learning capabilities demonstrated by these biological systems (Watkins, Timmis, and Boggess 2003; Timmis and Neal 2001; deCastro and Von Zuben 2002). A second appealing aspect of biological immune systems is their inherent distributedness. This paper examines these two strains by proposing modifications to an existing immuneinspired algorithm to provide a more efficient, parallel version. Since very little work has been undertaken, to date, in the parallelization of immune-inspired learning algorithms, this investigation is offered by way of a case-study in the gains paralleization can provide to current Artificial Immune Systems (AIS) and as a proof-of-concept in the field of parallel immune learning. The algorithm under investigation, CLONALG, utilizes the immunological concepts of memory cells, free antibodies, clonal selection, and affinity maturation to provide solutions for pattern recognition and function optimization tasks. This algorithm was developed by Leandro de Castro and offers us an ideal test bed for our initial explorations of immune learing in parallel. In (deCastro and Von Zuben 2002), the authors introduce an artificial immune system algorithm inspired primarily by the clonal selection theory first popularized by Burnet (1959). For a complete specification of this algorithm, please see (deCastro and Von Zuben 2002).