Artificial Intelligence research has become something of a tech arms race in recent years, with Google, Microsoft, and Facebook each running their own expensive cutting-edge labs. But today, Facebook's AI research lab (FAIR for short) is releasing new code for optimizing a certain kind of machine learning software. The new process can speed up machine learning algorithms as much as 23 times, according to Facebook's research, and as of this morning, it will be available for anyone who wants to use it. This probably won’t matter much for large labs at Google or Microsoft — but it may boost smaller apps or database projects that lack the resources or a larger company.
The difference between a pipe dream and a feasible feature
The code is specifically useful for convolutional neural nets, a kind of algorithm commonly used in image and video recognition. Facebook's new code would allow the same processing work to be done significantly faster or cheaper than previously available code: if sorting a given batch of images took a dollar’s worth of server time previously, the new code could perform it for less than 5 cents. For a startup trying to spot objects or faces in photos and videos, that could mean the difference between a pipe dream and a feasible feature.
Still, experts say the real significance isn't the research itself, but the open approach Facebook is taking to its research. "Whenever you're dealing with a for-profit lab, whether it's Google or Facebook, the question is to what extent will they be part of the academic community and, you know, play nice with others," says Oren Etzioni, CEO of the Allen Institute for Artificial Intelligence. "With this, they're putting a stake in the ground and saying, we are going to be part of the community."
Facebook founded its AI lab in 2013 with NYU's Yann LeCun, part of a larger movement that pulled machine-learning academics into the private sector. (Stanford's Andrew Ng took a similar post at the Google Brain project.) That's taken much of the cutting edge of artificial intelligence work out of the public sphere, as breakthroughs that would once be published in journals threaten to become trade secrets. "The purpose for Facebook is to facilitate communication between people, and that requires a pretty good model of what people want," LeCun told The Verge in 2013. "I have to believe that you need intelligent systems to do a good job on that."