MIT Researchers Use Ants to Study Social Networks
Biologists have long suspected that ants “size up” their population density by finding and bumping into each other.
The researchers think that this ability helps a colony of ants in finding and selecting a new nest.
More studies have been going into this theory.
Researchers from the Massachusetts Institute of Technology are going to present their findings in support of ant theory.
The university’s Computer Science and Artificial Intelligence Laboratory will be at a conference at the Association for Computing Machinery’s Symposium on Principles of Distributed Computing later this month to present their study, which shows that ants explore and observe their environment and come up with an estimate of their population.
This study can, according to the MIT researchers, help people analyze social networks.
One of the researchers, Cameron Musco, also a graduate student at the university’s electrical engineering and computer science department, said that the phenomenon in an ant colony can be translated into human communities this way:
The number of times people bump into each other (or how many people you bump into) gives you a good estimate of the number of people in the area. According to Musco, this can get better with time and with better equipment and with more study.
Sampling and Population
Together with his colleagues, Musco, portrayed the ant’s environment as a grid. The cells in the grid are occupied by a random scattering of other ants. The central ant starts with the closest cells, then with “equal probability” moves on to the next one.
With the same probability, the ant goes to another adjacent cell. This “random walk” in statistics means that the ant gets to count the number of other ants in the cells that it visits.
The researchers say the random walk is the random sampling during studies, and that the random walk gets results as quickly as random sampling. The accuracy of each approach improves over time.
This idea is important because in many cases, random sampling is not an option.
If someone wants to identify how many of Twitter’s members are Republican, it would be impossible as the researcher would require a list of all the members. The only way is to start with one member and look for its connections.
In ad hoc networks, similarly, a device can only pinpoint the location to the next nearest device.
Using the random walk approach, the researcher starts with one device or one Twitter member, then locates the nearest device or member. It goes on and on, until the researcher is able to gauge or estimate the population density of the sample.
One thing that the researchers noticed is that there is a likelihood for the starter ant or “explorer” to return to the cells it had already visited. This means that a random walk has more chances of oversampling than the random sampling procedure.
Musco and his teammates, Nancy Lynch, NEC Professor of Software Science and Engineering, and Hsin-Hao Su, a postdoc in Lynch’s group, realized that anything they did to change the algorithm of oversampling didn’t make things better.
But then they found out something else….
Musco says that in the entire grid, there are some ants distributed at the farthest areas. These ants are those that the explorer will never bump into, so, Mucso, reasoned, they have to count all their interactions with the nearer or “local” ants to make up for the faraway ants.
This approach, which can be polished and modified to use in other studies, makes the “random work” a very helpful tool in conducting researches on social networks.