Browse Books

Reviewer: H. Van Dyke Parunak

One of the most promising areas of heuristic computing is swarm intelligence, which imitates coordination mechanisms used by insects and other social animals. A hallmark of these mechanisms is stigmergy, the use of changes in a shared environment to convey information among agents. Ants provide the parade example of stigmergy in natural systems. They deposit chemicals, called pheromones, as they move about. Pheromones build up when several ants traverse the same area, thus reinforcing locations that are attractive to several ants. They evaporate over time, thus discarding obsolete information. They also propagate to nearby areas, generating gradients that can guide other ants to an area of high concentration. Ants use pheromones for many purposes, including finding paths joining their nests to food sources. Their success has inspired the use of digital imitations of pheromones for problems that can be mapped onto the path planning problem. Dorigo has been one of the most prolific exponents of this technique since 1992, refining it and extending it to a wide range of problems. This volume provides a comprehensive, integrated handbook to these techniques. Chapter 1 motivates the general approach, by discussing the behavior of natural ants, and exhibiting a simple algorithm that illustrates the main aspects of ant colony optimization (ACO). The algorithm differs in at least two important ways from natural pheromone-based optimization (and much other work in swarm intelligence). First, while relying on pheromone aggregation and evaporation, it makes no use of the propagation dynamic of natural pheromones. Second, it is a centralized, off-line optimization procedure that is applied to static problem instances (although some recent work, reported in chapter 7, has begun to relax this constraint). Chapter 2 generalizes the approach into a metaheuristic: a set of algorithmic concepts that can be specialized to address a wide range of potential applications. Chapter 3 describes a range of ACO algorithms that have been applied to the traveling salesman problem, a prototypical example of path optimization. For most of its history, ACO has relied more on experiment than on theory for an understanding of its behavior. Chapter 4 summarizes recent theorems, proving the convergence of two specific ACO algorithms, and lays the foundation for a more formal approach to the techniques. It includes a discussion of the insights into the functioning of ACO that are provided by the theoretical results that have been obtained. Chapter 5, the longest in the book, summarizes different applications of ACO that have been developed, including, not only routing problems, but assignment problems, scheduling problems, subset problems, and other nondeterministic polynomial (NP) hard and machine learning problems. This chapter discusses the general principles of applying ACO to real-world problems, and will be a rich source of inspiration for practitioners. Chapter 6 focuses on one specific application: packet routing in telecommunications networks. Chapter 7 offers a summary of the state of ACO research. It offers a glimpse of current trends, including applications of ACO to dynamic, stochastic, and multi-objective optimization problems, and includes a brief survey of other ant-inspired algorithms. The book will be invaluable to both researchers and practitioners in optimization theory and swarm intelligence. Each chapter includes a helpful summary, bibliographical notes, and exercises, making it suitable for use as a textbook. The book as a whole includes an exhaustive index, and a bibliography of over 500 references, through 2003. Online Computing Reviews Service