Building a more connected world: Facebook CTO
We all want to connect. No matter who we are or where we live, the one thing that unites us all is our deep desire to communicate.
At Facebook, we know how important it is for everyone to connect in a rich way, and we want to develop technologies to make it possible. As we look forward to the next 10 years of Facebook’s growth, three goals light our way:
Increased planetary connectivity
Today, fewer than half of the world’s 7.3 billion people are online, and 1.6 billion don't live within range of a data network at all. Through Facebook's Connectivity Lab we're working to bring connectivity to many who don't have access to internet infrastructure by developing technologies that bring the cost of connectivity down by an order of magnitude over current models.
The Lab's research is broad and the projects are wide-ranging. They include Aquila, our first high-altitude, long-endurance (HALE) unmanned aircraft system. When deployed, Aquila will be able to circle a remote region of the world for up to three months at a time, beaming connectivity down from 60,000 feet.
Today we announced two new terrestrial solutions we're pursuing, both breakthroughs in wireless technology. One is a network called Terragraph, which leverages unlicensed 60 GHz spectrum to create a high-bandwidth wireless solution for rapidly growing urban areas. The other, ARIES, is a proof-of-concept technology made of a new antenna array that has the potential to deliver 10 times as much information over the same bandwidth as 4G/LTE.
Facebook debuts terrestrial tech to deliver Internet
Better artificial intelligence
Developing high-capacity networks like these is important because the volume of data generated online every day is mind-blowing. More than 1 billion stories are posted across our Facebook apps and services daily; when people are not posting, they're watching more than 100 million hours of video every day and sharing more than 2 billion photos.
Powerful AI systems and the infrastructure to support them help us navigate this ocean of information. On the infrastructure side, we're scaling our software and hardware so we can do AI at massive scale, like power 6 million predictions per second. And we have grown our capacity to 40K TFLOPS per second available via the GPU server cluster used to train our deep learning systems. That will help us work smart systems into all of our apps and services.
While we scale our infrastructure, the Facebook AI Research (FAIR) team conducts promising research in several key areas, including computer vision, natural language understanding, planning, and reasoning. Even though the team was established just a couple of years ago, we've made important breakthroughs.
In natural language understanding, the part of conversation that's hardest for machines to understand is the unspoken context that often gives speech meaning. To address that, we have developed Memory Networks, or MemNets, which give AI the power to infer context based on its “memory” of previous parts of a conversation. We developed MemNets last year, and have been working this year to scale up the data set for MemNets while applying them in production. Leveraging MemNets, computers can understand language more like a human would.
Being open increases the pace of innovation, so we share our work with the community and stand on one another's shoulders to advance the field. In fact, we are a main contributor to Torch, an open source development environment for things like machine learning and computer vision, with a particular emphasis on deep learning and convolutional nets, which are huge areas of emphasis for FAIR.
AI is an important part of all of our futures, but integrating it at a billon-person scale presents significant challenges. So our Applied Machine Learning team is taking the full spectrum of topics related to AI theory, algorithms, and applications and bringing it to our products. We have more than 25 percent of our engineers integrating AI into our code in order to make your News Feed better, power new product experiences like Moments, protect your account, and keep unwanted content out of sight.
Integration of augmented and virtual reality
The best connection comes when people can be together even if they are physically very far apart, so we are developing virtual reality platforms, applications, and content that will help people share immersive new experiences, anywhere and anytime.
An exciting first step here is 360 video, which is ramping up very quickly. Millions of people are watching it on Facebook every day.
But the experience can get better and more immersive. Videos viewed in a VR headset can be 3D and more akin to how you actually experience the world. To aid in the development of that burgeoning ecosystem, we designed and built a high-quality 3D-360 video capture system, including camera hardware design and automated stitching software. Creating great 3D means capturing stereoscopic video. We get that using a rig that has 17 globally synced cameras: 14 of them bolted on a horizontal ring working together, with a single fish-eye camera on top and another two on the bottom. Their multiple images replicate the way your left eye and right eye work together to create a sense of depth. The stitching code works the rest of the magic, building on a concept called optical flow, where the system blends the images into one seamless panorama.
Virtual reality is a big part of our future. Both Gear VR and Rift from Oculus are available and off to an awesome start. More than 200 apps have already shipped for Gear VR, and people have watched more than 2 million hours of video. Rift just started shipping two weeks ago, but early reviews are very strong, and dozens of incredible games are available.
Facebook hires Google 'moonshot' exec for R&D
VR gives you a sense of immersion, but we're really interested in taking it one step further by building social presence. This requires customized hardware and software. It’s not just about one or two breakthroughs. Consider the core technology of Rift alone: high-quality displays with high refresh rates and pixel density; precise positional tracking; lightweight, sleek mechanics that are also adjustable; and cutting-edge optics. This includes technology advances required to get accurate avatars and body-language scanning so it can be represented in VR. And then there's what we interact with. Our hand controllers — Touch — will help in this effort. They add a real sense of presence as you're able to reach out and naturally interact with your environment, and thus able to interact with whomever you are playing with, no matter where they may be.
For newer platforms like VR, we're in the early parts of the accelerated growth curves. But we are heading in the right direction and are very excited about where we’re going.
These areas of research and engineering, and others like them across the Facebook family, fit together and support our mission: to give people the power to share and make the world more open and connected.
We will continue to dream and explore, and then we will build and innovate. We will publish our work, we will share our code, and we will collaborate with everyone in the industry.
I can’t wait to see what we’ll build — together.
Michael Schroepfer is chief technology officerfor Facebook. He contributed this guest column for USA TODAY.