by • July 13, 2016 • No Comments
Ants, the micro icons of industriousness and organization, apparently can tequite us a thing of how desktop networks work and how they can be improved as well. A team at MIT’s Computer Science and Artificial Intelligence Laboratory has been studying ants to turn it into a version of analysis of social networks, collective decision building one of robot swarms, and communication in decentralized, ad hoc wireless networks.
The MIT study ensures a long-held assumption in the scientific community, that ants estimate their population density based on the frequency at that they bump into other ants while randomly exploring their environs. This capacity seems to be key for several activities, which include deciding where the colony establishes a new nest.
“It’s intuitive that if a bunch of folks are randomly walking around an area, the number of times they bump into equite other can be a surrogate of the population density,” says Cameron Musco, co-author of a paper on the research. “What we’re doing is giving a complex analysis behind that intuition, and in addition saying that the estimate is a quite great estimate, fairly than a few coarse estimate.”
The researchers manufacture a parallel between an ant’s environment and a grid. An explorer ant starts at a few cell of the grid and most likely moves to one of the adjoining cells. It is in addition most likely that it and so moves to another cell adjoining to the one it departed of, and so on. In technical, statistical language this is called “random walk.” The explorer ant counts the number of other ants in all the cells it visits.
The study compares the random walk to random sampling, that is when cells are selected of the grid at random and the number of ants in equite cell counted. In both cases accuracy improves with equite extra
sample, but the surprising factor here is that the random walk can reveal the true population density approximately as rapidly as the tried-and-tested random sampling method does.
The researchers say this is relevant for the reason in most practical cases random sampling is not an version. For instance, in ad hoc networks, a given device knows only the locations of the devices in its immediate vicinity. An algorithm that uses random walks to aggregate information of multiple devices may be much simpler to implement than one that has to characterize the network as a whole.
The study and so gets informative for its additional counterintuitive findings.
It may be logical to expect that the explorer ant is most likely to return to a cell it has may already visited, so one given cell may additional most likely be oversampled than in the case of random samplings. But when the researchers got antsy of oversampled data and tried to filter it out, they discovered that, instead of improving their algorithm, it created it worse. They contribute a theoretical explanation for that.
“If you are randomly walking around a grid, you are not going to bump into equitebody, for the reason you are not going to cross the whole grid,” Musco says. “So there’s a fewbody on the far side of the grid that I have fairly much a zero percent accident of bumping into. But while I’ll bump into those guys less, I’ll bump into local guys additional. I require to count all my interactions with the local guys to manufacture up for the fact that there are these faraway guys that I’m never going to bump into. It sort of perfectly balances out.”
To version the ants’ environment, the researchers utilized a graph data structure consisting of nodes (circles), and edges, that are the line segments connecting nodes. In the grid, equite cell is a node, and it shares edges only with those cells immediately adjoining to it.
If the graph is not quite well connected, with only a chain of nodes, equite connected only to the two nodes adjoining to it, and so oversampling may become a problem, with the explorer stuck in the same set of nodes. But graphs describing communication networks frequently showcase two random walks starting of the same node and and so branching out in various directions. This being the case, random walks can contribute approximately the same level of accuracy as random sampling.
The bigger the number of explorers, the faster an analysis may create an accurate estimate. “If they were robots instead of ants, they may get gains by talking to equite other and saying, ‘Oh, this is my estimate’,'” Musco adds.
The researchers can present their paper at the Association for Computing Machinery’s Symposium on Principles of Distributed Computing conference, that can take place in Chicago between between July 25-29.
by admin • March 5, 2017
by admin • November 28, 2016
by admin • November 28, 2016