Swarm Intelligence is the discipline that primarily deals with natural and artificial systems constituted by different individual entities that co-ordinate using decentralized control and self-organisation. It mainly focuses on the collective behaviour of the individuals that result from local interaction among themselves and with the environment.
This field has marked multidisciplinary character since such systems can be observed in varied domains.
The basic properties of a Swarm Intelligence System are -
- It is composed of numerous individual entities
- The individuals are homogeneous
- The interactions among the individuals are totally based on simple behavioural rules that exploit only local information that is exchanged either directly or via the environment. (stigmergy)
- The individuals avoid nearby flock mates i.e. they always try to maintain a certain distance between themselves and others
- The group behaviour self-organises i.e. the overall behaviour of the group is a direct result of individual interaction.
The most fascinating dimension of a swarm intelligence system is the ability to act in a co-ordinated manner even without the presence of an external co-ordinator. Thus, even though no individual is in charge of a group, the group still shows overall intelligent behaviour.
While designing a swarm intelligence system, three characteristics must be maintained –
- Scalability: This means that the system can maintain high levels of functionality while increase in size without the need to train the individuals or redefine the rules of interaction. This is because the interactions involve only neighbouring individuals, thus increase in the number of individuals does not proportionally increase the number of interactions. In artificial systems, this is extremely significant since a scalable system can increase its performance by simply increasing the size without having the need to reprogram.
- Parallel functionality: This is possible in a swarm system as individual entities can perform different operations at different places at the same time. This helps in making an artificial swarm intelligence system more flexible, and enables it to efficiently self-organise and perform different aspects of a highly complex task.
- Fault tolerance ability: This is an inherent property of swarm intelligent systems due to their decentralized nature and self-organising control structures. Since the system is composed of different individuals none of which are solely important to the overall system behaviour, a faulty individual can be easily replaced by an efficient one.
Examples of swarm intelligent systems can be found widely in nature. Clustering behaviour of ants, nest building behaviour of termites, foraging pattern of honey bees, flocking of birds – all are highly efficient swarm intelligent systems showing efficient co-ordinated group behaviour. The following section briefly describes a few such systems.
- Clustering behaviour of ants
This is one of the most easily observable examples of swarm intelligence in nature. Ants build cemeteries by collecting dead bodies. They also organise the spatial disposition of larvae into clusters in a way that the younger ones are placed at the centre and the older ones at the periphery. This clustering behaviour has been a topic of much scientific study and experiments. Researchers have been able to build a probabilistic model of such behaviours and have obtained positive test results during simulation. They have also been validated by experimental data obtained with real ants.
- Nest building behaviour of termites
The mounds build by termites are enormous with their dimensions reaching many meters. However, compared to an individual termite, this is more than a thousand times its size. Scientists have been studying the co-ordination mechanism of termites, which allows them to construct such amazing structures and have come up with probabilistic models exploring stigmeric communication between the individuals to explain the behaviour. Some of these models have been implemented in computer programs to produce simulated structures corresponding to such nests.
Similar activity can be witnessed in wasps as well who construct nests having an extremely complex internal structure which is well beyond the cognitive capabilities of a single wasp.
- Flocking of birds
Flocking of birds is an example of well co-ordinated group behaviour. Researchers have proved that such kinds of refined swarm-level behaviours can be thought of as a result of self-organised behaviour where no individual acts as the controller and each of them base its movement only on locally available information, gathered from its surroundings. These studies have inspired a number of computer simulations like Boids simulation program. They are primarily used in the computer graphics industry for the realistic reproduction of flocking in movies and computer games.
Modern day technology is being employed to design and build swarm intelligent systems to solve a variety of different problems.
Ant colony optimization and particle swarm optimization are two such population based methods to determine approximate solutions to difficult optimization problems.
- Ant colony optimization
This type of optimization technique is inspired by the foraging behaviour of ant colonies. In an ant colony optimization, a set of artificial ants search for good solutions to an optimization problem which is transformed into the problem of finding a minimum cost path on a weighted graph. The ants incrementally build solutions by moving around in the graph. This approach has numerous real world applications. It has been used to optimize routing in communication networks and also in probabilistic travelling salesman problem.
- Particle swarm optimization
This again draws inspiration from the group behaviour of birds and fishes. In this type of optimization, the particles search for good solutions to a continuous optimization problem. Each particle derives its own solution based on its experience as well as that of the neighbouring particles, similar to flocking behaviour of birds. This optimization technique is also successful in solving different problems like pattern recognition, image processing etc.
The engineering based application of swarm intelligent systems can be found in network management, swarm based data mining, clustering, job scheduling and in robotics to instil cooperative behaviour in swarms of robots.
- Swarm based Network Management
Swarm intelligence is applied in network management to enable efficient routing and load balancing in communication networks. Ant-based Control (ABC) and AntNet are two such algorithms developed for this reason. Their performances were found to be very much stable and efficient even during instances where high traffic was observed.
- Swarm based Data Mining
Data mining and particle swarm optimization may seem that they do not have many properties in common. However, they can be used together to form a method which often leads to the result, even when other methods would be too expensive or difficult to implement. Particle swarm optimization is used to solve optimization problems in the construction of virtual reality spaces for the representation of data and knowledge in visual data mining. Even artificial ants are trained to pick up and drop data items with probabilities that are governed by the similarities of other data items already present in their neighbourhood.
- Co-operative behaviour in Robots
Swarm robotics is an innovative approach in solving problems by using a team of robots. Such robots show clustering behaviour and efficient co-ordination in tackling real world situations.
Swarm Intelligence can also be used to almost accurately render crowd simulation and group behaviour. Numerous algorithms have been developed to imitate such co-ordinated group behaviour to be used in problem solving. Some of them are artificial bee colony algorithm, cuckoo search, firefly algorithm, gravitational search algorithm, intelligent water drops algorithm etc.
Swarm intelligence is thus regarded as one of the most important technological advancements in recent times which, if efficiently applied to model systems can change lives in future.