Facebook's quest to get the world online is paying some unexpected dividends. Its Connectivity Lab is using image recognition technology to create population density maps that are much more accurate (to within 10m) than previous data sets -- where earlier examples are little more than blobs, Facebook shows even the finer aspects of individual neighborhoods. The trick was to modify the internet giant's existing neural network so that it could quickly determine whether or not buildings are present in satellite images. Instead of spending ages mapping every last corner of the globe, Facebook only had to train its network on 8,000 images and set it loose.
The effort isn't strictly altruistic. More detailed maps let Facebook know which areas are most likely to need better internet access, and which technologies (such as aerial drones) might work best. It could help tailor emergency responses, too. However, the social network isn't hoarding the data for itself. It's partnering with Columbia University to create a public data set that should be available later this year, making sure that businesses, governments, scientists and the curious are better-informed about where people live.