Out in the Open: Free Software That Teaches Your Smartphone How to See

Jetpac

Pete Warden has been trying to teach computers to see since the 1990s. Now, thanks to the branch of artificial intelligence called deep learning, he’s finally making some progress.

Deep learning attempts to model the structure and behavior of the human brain to solve complex computer science problems. The field has been around since the 1980s, but there’s been an explosion of interest in its techniques in the past few years as the cost of powerful computers has fallen. Google now uses deep learning inside several of its online services, and last year, it hired Geoffrey Hinton, the central figure in the movement. Not long after Facebook hired his colleague Yann LeCun.

Warden, an ex-Apple engineer, studied machine vision as an undergraduate, and recently, he adopted deep learning at his startup Jetpac, a company that analyzes Instagram photos in order to make travel recommendations. It has worked so well, he’s now sharing his methods with the world at large, offering a free developer kit called DeepBelief. Able to run on smartphones, tablets, even the tiny Raspberry Pi computer, the DeepBelief framework aims to provide all programmers with a simple way of building mobile apps that can recognize simple images.

Jetpac offers the developer kit under an open source license, so anyone can use it to build applications for any purpose. “We’ve benefited from the hard work of Geoff Hinton and other pioneers,” he says. “So I wanted to give something back in return for all of the awesome research that they’ve contributed.”