Data mining has changed the way we think about information. Machine-learning algorithms now routinely chomp their way through data sets of Twitter conversations, travel patterns, phone calls, and health records, to name just a few. And the insights this brings is dramatically improving our understanding of communication, travel, health, and so on.
But there is another historical data set that has been largely ignored by the data-mining community—photographs. This presents a more complex challenge.
For a start, the data set is vast, spanning 150 years since the dawn of photography. What’s more, the information it contains can be hard to distill, often because it is too complex or too mundane to describe in words.
Today, that changes thanks to the work of Shiry Ginosar at the University of California, Berkeley, and a few pals, who have pioneered a machine-vision approach to mining the data in ordinary photographs.