How do you count people? It’s not as easy as you might think.
You might try looking at birth and death records, or population censuses – but in many African countries (and, indeed, countries around the world) those records may not exist or, at least, may not be very accurate.
If you’re Facebook, trying to figure out where people live and what geographic challenges prevent them accessing internet services (and becoming users, obviously), involves looking at pictures. Lots of pictures.
The firm has just announced the first results of a collaboration with Columbia University and the World Bank in which its data science team has been looking at satellite imagery to work out where people live, and thus infer ways to connect them.
It’s taken a sample of imagery covering 23 countries, and by training an AI to recognise buildings and the likelihood of them being inhabited, then cross-referencing its findings with known population data, it reckons its got to within 6% of real world measurements with a system that can be deployed globally.
Writing on the Facebook blog, Tobias Tiecke says that one challenge was working out that buildings built for similar purposes look different in different parts of the world. Or, as he puts it:
“We used state-of-the-art convolutional neural networks to develop a model that has the accuracy to individual buildings and simultaneously works on satellite images across the globe. In order to achieve these contradicting requirements, we had to make tradeoffs in the performance of the model. There has been a lot of work recently on neural networks that can recognize individual buildings with very high accuracy, but these models are finely tuned on the local characteristics of the region where they are trained. We found that these models do not perform well at a global scale with realistic amounts of training data.”
That problem overcome, there’s some interesting findings from the initial analysis. For a start, says Tiecke, just under half the population in the countries examined live in cities already, and 99% live withing 63km of a city.
That’s relevant because “If we are able to develop communication technologies that can bridge 63 km with sufficiently high data rates, we should be able to connect 99% of the population in these 23 countries.” While such a technology does not yet exist, it will inform how Facebook might use its Aquila drones or internet.org satellites to bounce internet signals around.
If you want to see more, the first five data-sets have been made public at CIESIN at Columbia University and cover South Africa, Malawi, Ghana, Haiti and Sri Lanka. Tiecke says more data will be released in due course.
We’ll have more once we’ve downloaded the half a gig ZA zip file.
[Main image – Pretora population count as seen by CIESIN]