The sight of someone tracing the footsteps of tiny ants may seem absurd. But many researchers did carry out this painstaking exercise, only to be hugely rewarded. For instance, today’s communication and network engineers have picked up a trick or two from these social insects and are using them to manage the ever-increasing telecom and Internet traffic.
It all began in the late 1980s and the early 1990s. Inspired by the incredible communication skills of ants, researchers at Hewlett Packard, Bristol, Delft University of Technology, The Netherlands, and the University of West England, UK, decided to apply some of the lessons they learnt from worker ants in their businesses.
Once a worker ant discovers a food source or a new nesting place, it leaves a message as a trail for others. The message is in the form of chemicals called pheromones, synthesised in specialised exocrine glands and secreted directly to the external environment. Ants encountering the trail receive the message through chemoreceptors located in their antennae. Pheromones are volatile and evaporate easily, thus ensuring an upward movement from the trail to the receiving ant’s antennae. But because of the volatile nature of the chemical, the trial dissipates within two minutes. A worker ant can travel about 40 cm in two minutes — so if a fresh trail is to be encountered, another foraging worker must cross the trail or be directed to it by tactile communication. The second ant hitting the trail, however, reinforces it by adding more pheromones, making the trail more prominent and others to follow would also do the same. Once the food is exhausted, no more pheromones are secreted.
What interested the researchers was the ability of ants to find the shortest path to a food source via a process of self-organisation, mediated through pheromone trails. The ants also remain adaptive, that is, if the environmental condition changes — for example, a path is no longer available — they find a quick solution.
According to the researchers, these concepts could be applied to addressing problems of load balancing and message routing in telecommunications networks. They developed “electronic ants” with the ability to deposit “electronic pheromones” while moving along a telecom network. A major problem of telecom networks is the overcrowding of calls on a particular node. This could lead to loss of calls, that is, failure to get connected. Ants quickly respond to difficult situations — by adjusting the pheromone secretion, they can change their route as and when required. The objective of the artificial ants is the same — they would be able to select the appropriate nodes so there wouldn’t be overcrowding anywhere.
The collective problem-solving capability of ants is described as “Swarm Intelligence (SI),” a term coined by computer engineers G. Beni and J. Wang in 1989. A typical SI system is composed of a population of agents interacting locally with one another and their environment. Despite the absence of a central coordinating structure, the local interactions help the emergence of quick and advantageous behaviours for all the members. In the living world, SI is observed in ants, honey bees, wasps, termites, flocking birds and schooling fishes. The weaknesses of the individual brains are compensated for and this adds greatly to the survival of these lower organisms. In the case of ants, the added advantage is called “Colony Optimisation”.
The biological agents in an SI system could be replaced with artificial agents, like the “electronic ants” mentioned earlier. This would be in the realm of Artificial Intelligence. Now SI, especially that of ants, is very much a part of Artificial Intelligence. In recent years SI, especially Ant Colony Optimisation, has been eliciting tremendous interest from electronic and IT researchers. In 1984, an international workshop on Ant Colony Optimisation and Swarm Intelligence was held at Brussels, Belgium. The workshop discussed various aspects of SI, like models that can stimulate new algorithmic approaches, empirical and theoretical research in ant colony optimisation and SI, application of ant colony optimisation and SI methods to real world problems and also swarm robotics. Very recently, in July 2006, a special session on SI was held in connection with the Congress on Evolutionary Computation held at Vancouver, Canada.
With the explosive growth of the world wide web, it is becoming increasingly difficult for websites to attract visitors. Learning the patterns of navigation might help in the reorganisation of websites according to the representative surfing models, automatic prediction of URLs of interest for web users and pre loading of web pages in the server memory to speed up connection time.
Researchers at the Université de Tours, France, recently came up with a novel method for doing exactly that. Again, they were inspired by the chemical communication of ants. Ants manage to identify intruders by a colonial odour established through chemical exchanges between nest mates. The researchers developed a programme called “AntClust” imitating ants’ cuticular odour, their recognition template and genome. AntClust associates one object of the data set to the genome of an artificial ant. It also mimics the acceptance scheme between two ants and reproduces behavioural rules that allow the algorithm to converge. The cuticular odour of each ant is coded as a number representative of its nest. The behavioural rules modify the label of each ant until it finds a nest where it is well integrated. AntClust manages to generate meaningful clusters from sessions coded with “hits by page” that help in interpreting the users’ interests. The researchers now plan to evaluate AntClust with larger data sets and several websites.
During the 3rd Canadian Conference on Computer and Robot Vision, 2006, Canadian researchers Alice R. Malisa and Hamid R. Tizhoosh of the University of Waterloo presented the results of their studies using Ant Colony Optimisation for image “thresholding”. Based on their experience, the researchers proposed an approach where one ant is assigned to each pixel of an image and then moves around the image seeking low gray scale regions. They further claimed that their model performed better than the other two established “thresholding” algorithms now available. However, the technology is yet to come out of the laboratory walls.
(Dr Sumodan is a lecturer in Zoology, Government College Madappally, Calicut, Kerala.)
Graphic: Gargyee Bhattacharyya Roy