4,000 million people have no postal address and 'machine learning' tries to correct it | Innovation
Some 4,000 million people lack a physical address. It is not a minor issue: residents lose access to important services such as package delivery, medical care and disaster assistance, as well as the ability to register to vote or get a driver's license. Cities also have problems planning new infrastructures, such as schools, water pipes and power lines. The MIT Media Lab and Facebook propose a new way Direct address to those who do not have it with automatic learning.
The researchers, according to account the MIT Technology Review, They established two teams: a first trained a deep learning algorithm to extract the road pixels from the satellite images. Another algorithm grouped these pixels and joined them together in a road network. The system analyzed the density and shape of roads to segment the network in different communities.
The densest group was labeled as the center of the city. The regions around the center of the city were divided into north, south, east and west quadrants, and the streets were numbered and put letters according to their orientation and distance from the center.
How accurate are the machines? When they compared their final results with a random sample of unmapped regions whose streets had been manually labeled, their approach successfully addressed more than 80% of the populated areas. Much? Little bit? This data improves the coverage of Google Maps or OpenStreetMaps, according to the MIT.
There are other ways to automate the creation of addresses: The start what3words generates a unique combination of three words for each square of three by three meters in a global grid. The plan has already been adopted in the regions of South Africa, Turkey and Mongolia by national parcel delivery services, local hospitals and regional security teams.
Ilke Demir, Facebook researcher and one of the creators of the new system, assured the US publication that its main advantage is that it follows the existing road topology and thus helps residents understand how the two directions relate to each other, not as It happens with what3words. "If a person has the address done with three random words like parrot, failure, casino and another has the address mesa.silla.televisión, you will never know if it is a neighbor of that person, "he says." That is the key: we want directions that people can intuitively relate. "
