Facebook’s French offices in Paris’s 7th arrondissement, normally filled with sales and marketing folks, are getting some new tenants this week. Yann LeCun, the Facebook scientist responsible for some of the most cutting-edge artificial-intelligence projects around, is opening a new lab in Paris that’s slated to include approximately 12 researchers. The ambitious project is an example of Facebook’s dedication to becoming one of the world’s top companies in automatically recognizing faces in pictures or teaching computers to understand human speech: Instead of getting Europe’s top talent to come to America, Facebook is making the trip in reverse. The lab is slated to start operations on Tuesday, June 2, and LeCun will remain in New York overseeing research in New York, Paris, and California.
“The reason we’re doing the lab in Paris is to take advantage of the pool of talent in Europe,” LeCun told me when we hopped on the phone Monday morning. “The highest concentration of experts in Europe is in Paris and London. Zurich is third, then there are lower-density regions in Germany, Spain, and Italy.” LeCun cited the Xerox Research Centre in the French city of Grenoble, which is doing cutting-edge machine vision research, as an example of France’s pull when it comes to machine learning and deep learning.
LeCun, who was poached by Facebook from academic work at NYU several years ago, is one of the world’s top experts the field of “deep learning.” Deep learning basically involves teaching computers to understand, extrapolate, and make inferences from extremely complicated data like photographs, technical manuals, and and voice recording. It’s also extremely important to the future of the world’s biggest tech companies. Google lured Geoff Hinton, another of the field’s top experts, away from academia to work on projects.
Facebook has also aggressively sought to open its deep-learning and machine-learning projects up to the general public and to academia because, as LeCun says, corporations and academia “need each other.” Earlier this year, Facebook made some of its deep-learning tools available to the public, and (just like Google) publishes research papers on a regular basis.
LeCun and Hinton recently collaborated with another expert, Yoshua Bengio of the Universite de Montreal, to publish a research paper on the state of deep-learning research in Nature. In the paper, which is aimed at computer science experts, the trio explain the technique’s applications for teaching computers to learn information from images, video, audio, and speech. While at the moment, deep learning is used by Facebook for techniques such as facial identification in images, it can be used for much more. And Facebook intends to be at the forefront of corporations adept at deploying deep-learning techniques. As LeCun told me, it has big ramifications for any institution–from law enforcement to health care–that uses images in the course of their work.
“If you have imaging data or anything else that needs analysis, it will be easier for you to find techniques to apply to your program. There are areas like medical imaging which have a lot of applications for medical diagnosis–broadly, it is is a very positive development,” he told me. “There is the use of deep learning for cars for driving safety. There are uses such as detecting pedestrians, collision avoidance, or even self-driving cars in the future. There is Mobileye, the Israeli company working on machine vision for cars, Nvidia’s big project to do the same thing, and Uber’s research lab in Pittsburgh near Carnegie Mellon University. There are a lot of people in a lot of domains looking at a lot of applications.”
Facebook’s Paris lab will initially contain six researchers; the company says by the end of 2015, there will be 12 researchers. The company’s offices in Paris mainly hold marketing and sales staff.