Owen Holland, University of the West of England, Bristol BS16 1QY, UK
This paper sets out to characterise some aspects of the new wave of ant based computations, and to locate them within a larger space of related methods and problem domains. We begin by giving an overview of the recent burst of activity in this area, noting the type of ant phenomena modelled, and the typical problem domains. We then identify some key features of ant based methods, and the problems to which they are applied, in a quite general way, and examine some current and historical examples of techniques based on electricity, chemistry, and simple automata which can be seen to be related to ant-based methods. This is followed by a consideration of the basic abstract abilities of real ants, including some factors which are currently used in computational models, and some which have not yet found application; the inappropriateness of using some standard agent models in the context of ant behaviour is discussed. We present a number of new studies showing how certain aspects of ant behaviour can be used to solve some primitive abstract spatial problems, which may well underly the success of ant algorithms in more complex problem areas. The paper concludes with a discussion of one particular application of ant based techniques [Schoonderwoerd et al - the SHBR model] to the problem of load balancing in telecommunications networks, and presents new simulation results and related work intended to deepen our understanding of how the technique works.