Day 2 of F8 2018: Developer News Roundup

Day 1 of F8 was packed full of important developer news, including details on our enhanced app review process, the release of Graph API 3.0, and several new products — check out our F8 Day 1 Developer Roundup to dig into all the updates. Day 2 of F8 continued the momentum, especially on the important topic of artificial intelligence. Here are just a few of the highlights from today's AI updates:

PyTorch 1.0, New AI Platform for Research to Production

To help accelerate the path from AI research to production, today we announced the next version of our open source AI framework, PyTorch 1.0. With PyTorch 1.0, AI developers can seamlessly transition between a flexible, immediate execution mode for research, and a highly optimizable, graph execution mode for production. PyTorch 1.0 will be available in beta in the coming months. You can learn more about all of our AI tools at facebook.ai/developers. Read more

ONNX Expansion Speeds AI Development

Facebook helped develop the Open Neural Network Exchange (ONNX) format to allow AI engineers to more easily move models between frameworks without having to do resource-intensive custom engineering. Today, we're sharing that ONNX is adding support for additional AI tools, including Baidu's PaddlePaddle platform, and Qualcomm SNPE. ONNX is also adding a production-ready converter for Apple Core ML technology. With these additions, ONNX now works with the vast majority of model types and can be deployed to millions of mobile devices. Read more

Advancements in Computer Vision

Because image recognition and understanding is such an important field in AI, our Applied Machine Learning and Facebook Artificial Intelligence Research (FAIR) teams have trained a new computer vision system using an unprecedented 3.5 billion publicly available photos, using public hashtags to classify the images. In his keynote earlier today, Srinivas Narayanan, director of Facebook's Applied Machine Learning group, described how a 1 billion-image version of this data set enabled our image recognition tool to score the highest mark ever on the widely used ImageNet benchmark — 85.4 percent accuracy. Read more

Open Sourcing Code for AI Research Community

The FAIR team has developed sophisticated game-playing AI bots for the board game Go and coming soon, for the real-time strategy game StarCraft. To help the larger AI research community, we will continue to open source code as well as the data sets and models for as much of our work as possible. We're also sharing FAIR’s work in partnership with university researchers to create AI tools that learn to navigate indoor spaces and use natural language to answer questions. We've open-sourced data from this project as well, including House3D, a collection of virtual environments to train agents. Read more

For more Facebook news and announcements from F8, check out the Facebook Newsroom. You can also watch the F8 keynotes and sessions.