In 1995, James Kennedy and Russel Eberhart published their seminal work on swarm intelligence (incidentally, their work provides an interesting lesson in cross-disciplinary collaboration; Kennedy is a specialist in social behavior of animals, and Eberhart is an electrical engineer). They described the use of a new computing technique, which they called particle swarm optimization (PSO), to develop empirical mathematical models known as neural networks (these, in turn, are patterned after the organization of neurons in brains). The technique was developed based on early attempts to simulate life-like movement of virtual animals in virtual space. The end result was an algorithm wherein "agents" or particles fly through multidimensional virtual space in search of a "best location" corresponding to the optimal solution to a problem; the flight of the particles is continually adjusted based on simple memory and information-sharing rules within the swarm. (The website http://www.particleswarm.info provides plenty of information for both novices and specialists). PSO turned out to be a very robust, effective algorithm for various problems, and in the past decade it has been used with varying degrees of success for applications that include optimal scheduling, engineering design, diagnosis, data mining and forecasting.
PSO is only one of several algorithms that mimic the behavior of social animals. Other techniques based on ants, frogs and bees have been developed, and they all fall under the general heading of swarm intelligence. Swarm intelligence, in turn, falls under a more general class of algorithms, known as metaheurisitics, which includes evolutionary and genetic algorithms as well as simulated annealing. These computing techniques combine adaptive behavior with random influences in order to facilitate the search for optimal solutions to problems, and to minimize the risk of getting trapped into deceptive local solutions.
My own experience with PSO dates back to 2003, when I had the naive notion that it would be a magic bullet of sorts to solve difficult design problems in my field of process integration. The first program I wrote was a complete failure. Fortunately, I have since had increased success with the method for designing clean, environment-friendly industrial processes – for example, choosing the best combination of pollution management techniques to implement in an industrial plant, or designing the best scheme for recycling water through a network of pipes to reduce waste generation. Two of my graduate students, Hul Seingheng and Kristine Col-long, spent about a year each working on variants of the latter problem. The work has been the subject of both collaboration and friendly competition with other researchers in Malaysia, Cambodia, United Kingdom, Japan and the United States.
The application of swarm intelligence to solve environmental problems is a fast-moving field. The first demonstrated use for designing industrial water conservation schemes was published in mid-2006 by a research group in China – two or three months ahead of the work of my own team, much to our dismay. The potential for development seems limited only by what can be imagined. Other applications that we are working on at the Center for Engineering and Sustainable Development Research (CESDR) in La Salle include designing eco-industrial networks, planning renewable energy systems, and predicting the properties of biodiesel from its chemical make-up. Other potential applications I think should be explored include medical diagnosis and traffic routing to alleviate urban congestion. At the same time, there are many research opportunities in developing enhanced variants of PSO, for example, by developing hybrid techniques incorporating features borrowed from other algorithms. Our own experience in La Salle has demonstrated the effectiveness of utilizing random mutation events and "population wipe-outs" for some engineering applications. Eventually, it may be possible to develop a uniquely homegrown algorithm, one inspired by direct observation of an as yet unnoticed natural phenomenon.
Looking at the whole concept of swarm intelligence from an entirely different angle, we might ask the question: Can we draw any lessons from these virtual swarms? In fact I recently read a PSO paper by Mendes and Neves with the provocative title, "What Makes a Successful Society?" which hints at how swarm dynamics can give us insights on how real agents (e.g., people) behave under different conditions, and on how these components, in turn, determine the ability of groups of individuals to progress. One of the fundamental lessons is that the group is smarter than the sum of its parts – apparently there really is such a thing as collective intelligence, at least in the world of the virtual swarms. The particles that make up a typical swarm are very simple entities, and the apparently purposeful behavior exhibited by the swarms emerges out of group dynamics through information exchange. On the other hand, it is quite clear, both from the swarm literature and from my own experience, that one of the main reasons swarms stagnate short of their ultimate goal is the loss of diversity. In our work in La Salle, we have tried different strategies to preserve swarm diversity in our programs – using algorithmic operators with colorful names such as "mutation" and "turbulence." I am thus strongly inclined to believe that in any real-life enterprise, it is just as essential to always maintain a healthy level of diversity of thought, in order to maintain a solid foundation for future progress.