Computer Vision News - November 2018

7 Computer Vision News also. Later, my reading matter was people like Geoff Hinton. I worked with Geoff as a postdoc and discovered we had a lot of common interest. Let’s talk about Facebook. You are hiring a lot of the best software talent that is available today in the artificial intelligence world. I’ve interviewed many of them and there are many more who are working for Facebook now and for some of the other major corporations. What is funny is that most of them don’t work in the core business of Facebook, but in something that is important for you at the mid to long-term. Without asking you anything confidential, what can you tell us about this? We don’t have anything that’s confidential. Well, not many things are confidential, because Facebook AI Research is the fundamental research lab in AI at Facebook and it’s outward facing. It’s very much connected with the research community. We publish everything we do. We distribute a lot of our code in open source. We collaborate with universities. We have interns and resident PhD students in France and the US. It’s very open. This is good in general, and of course, it will be good for Facebook in the long-term, because the main limitation of AI today is not whether Facebook is ahead of Google or Microsoft or IBM, it’s rather that the field itself is not where we want it to be. If you want to build intelligent virtual assistants, for example, then the science or the technology to build intelligent machines that have a bit of common sense to interact with humans does not exist. So our objective is to develop the techniques that will make that product a reality. We don’t have a monopoly on good ideas, despite the fact that we’re hiring a lot of the top people. So, we have to interact with the broader research community. That’s why we are open. There’s also another set of organisations within Facebook, part of the broader “Facebook AI” organisation, which are much more focused on problems related to Facebook – computer vision, natural language processing, search, things like that. A lot of those groups use technologies that were originally developed at FAIR, maybe for a different purpose, and so there is quite a lot of influence there, but these groups have a different modus operandi. They are much more focused on the needs of the company. They publish papers also, but not as much. They are focused on improving the services that Facebook provides or creating new ones. FAIR is working on developing new technology and advancing the field. Sometimes we think the horizon is 3 years or 5 years or 10 years, but sometimes, the things we come up with turn out to be useful right away. It surprises us sometimes. In your opinion, what is the best computer vision paper of 2018? I’m sure you won’t be surprised to hear that our magazine named Mask R-CNN as the best paper of 2017. I would not disagree either! [ both laugh ] So many things are happening “We don’t have a monopoly on good ideas, despite the fact that we’re hiring a lot of the top people!” Yann LeCun Guest