A team of Facebook engineers has spent their last two years designing and developing drones, satellites, and lasers that could bring Internet access even to the people in the most remote areas on Earth.
However, all of this work is useless the social network can come up with a way to locate those people – for which a little artificial intelligence is needed.
Therefore, Facebook is now using its impressive computing muscle to create maps of inhabited places and the way location affects internet connections. Eventually, these maps should enable Facebook to determine which solutions work best for which types of connectivity.
All this work is part of Facebook’s Connectivity Labs, the technical arm of its Internet.org initiative. This department deals with satellites, drones, and lasers for providing internet to rural areas, emerging markets and developing countries.
With better maps comes better Wi-Fi hotspots or cellular technologies, because Facebook will be able to better assess the needs of the people based on their location.
In their effort of generating the maps, the engineers in the Connectivity Labs partnered with Facebook’s data infrastructure unit and machine learning, science division, and artificial intelligence groups.
This giant coalition of teams then analyzed tons of satellite data covering 21.6 million square kilometers in 20 countries; over 350TB of data were processed.
They used a combination of computer vision techniques – such as the image-recognition algorithm Facebook uses to identify people’s faces in photos – in order to identity human-built structures.
It’s worth mentioning that the company was adamant about the fact that no Facebook photos were used to create its data maps. With the help of AI techniques and by applying machine learning, Facebook managed to hone the maps.
In the end, the engine could successfully identify “outlines of buildings and highlighted those for which it had high confidence while suppressing areas not likely to contain human-made structures.”
Once the structures were detected, Facebook could use them as proxies for where people lived. By cross-referencing these maps with census data records, the coalition was able to distribute the population evenly across each location.
This is in no way an error-free system, but according to Facebook, the end result used the method that gave “the least error-prone way of determining how many people lived in each building.”
Sometime later this year, Facebook will be releasing its data to the general public, thinking that their research could have many more impacting applications, such as risk assessment for natural disasters.
Image Source: Gigaom