Neighbourhood watch: A machine-learning census of America’s cities

Neighbourhood watch
A machine-learning census of America’s cities
from The Economist

In 1686, Louis XIV came up with an idea to take a census and count France’s resources to control its towns and nobles better

Such surveys are expensive. But a team in Stanford came up with a cheaper, quicker method by using ML algorithms & data collected by Google

Using data from automotive websites, they trained ML algorithms which spotted 22m different cars on 50m images from Google Street View

Correlating the cars with data from traditional census, ML algorithm predicts education levels or political leanings from cars in an area

– The more detailed predictions are the less certain they become. As predictions rely on traditional surveys, it’s unlikely to replace them

+ The ML system is much cheaper and faster. It crunches through 50m images in two weeks. A human would take 15 years to do the same

+ Self-driving cars or Earth-imaging satellites produce even bigger data sets. This ML system could soon become a constantly updated one