|Facebook Advances in AI At F8|
|Written by Sue Gee|
|Friday, 04 May 2018|
Facebook is one of the big technology companies that is making a big investment in AI and at F8 it was ready to share some of its latest progress. The good news is that Facebook is open-sourcing its AI tools on GitHub and there is a new website for accessing them.
Attendees at F8 were the first to learn that Facebook's OpenGo bot has achieved professional status after winning all 14 games it played against a group of top 30 human Go players. Facebook CTO Mike Schroepfer obviously wanted to view this success in the context of Google's AlphaGo by saying:
"We salute our friends at DeepMind for doing awesome work."
As Facebook notes on its blog post today, the DeepMind model itself also remains under wraps. In contrast, Facebook has open-sourced its bot.
"To make this work both reproducible and available to AI researchers around the world, we created an open source Go bot, called ELF OpenGo, that performs well enough to answer some of the key questions unanswered by AlphaGo."
Created by the FAIR (Facebook AI Reaserch) Team ELF, an Extensive, Lightweight, Flexible platform for games research which allows researchers to test their algorithms in various game environments, was open sourced last year. ELF runs on a on a laptop with GPU, and supports training AI in more complicated game environments, such as real-time strategy games, in just one day using only six CPUs and one GPU. It’s not just Go that the FAIR team is interested in. It has also developed a StarCraft bot that can handle the often chaotic environment of that game which is plans to open-source.
A new version of Facebook's open source AI Framework was also announced. The PyTorch framework has quickly become one of the most popular frameworks for AI researchers. Since Facebook released the original version (0.1.6) just over a year ago, it has taken off on GitHub, and was the second-most cited framework in papers at ICLR. PyTorch 1.0 will be in beta in the coming months.
Facebook also announced the expansion of ONNX (Open Neural Network Exchange) which as we reported when it was released was co-developed by Facebook and Microsoft. ONNX is an open format for representing deep learning models, to allow AI engineers to more easily move models between frameworks without having to do resource-intensive custom engineering and is tightly woven into PyTorch 1.0.
In addition Facebook is releasing resources for more specific uses of AI. Its ResNext3D model, with state-of-the-art accuracy and efficiency for video understanding, will be available in June; Translate, its PyTorch language library for machine translations that scale, and sharing our early work on a project called Multilingual Unsupervised and Supervised Embeddings (MUSE), to help expand the number of languages available for translation.
The above links go to a new developer-oriented website, Facebook.ai, which:
is intended to help researchers and engineers outside of Facebook take their AI work from research to production more quickly, employing the exact same tools that we use to serve 2 billion people. By sharing this growing suite of open source tools, we're underscoring our commitment to work collaboratively with the developer community to build meaningful new products and services.
To return to the advances announced at F8, Facebook reported the news that a collaborative effort by the Applied Machine Learning (AML) and Facebook AI Research (FAIR) teams has achieved a breakthrough in computer vision. A new image recognition model has been trained on an unprecedented 3.5 billion publicly available photos, using hashtags to classify the images. The researchers had to create a new technique that makes hashtags, which are noisy and imprecise labels, useful for AI training. Using a 1-billion-image version of this data set enabled the model to score the highest mark ever, 85.4% accuracy, on the ImageNet image recognition benchmark.
Commenting on its success Facebook says:
This performance reveals the long-term potential of not only improving image recognition by training on larger data sets, but also using existing labels (rather than annotations applied specifically for the purposes of AI training). We plan to open-source the embeddings of these models in the near future.
In the field of 3D image mapping, Facebook researchers have used the PyTorch toolkit to generate full 3D surfaces that can be applied, in real time, to footage of human bodies in motion. The new tool, DensePose, which Facebook hopes to publish as an open source library in the coming weeks:
.can enable compelling new augmented reality effects, such as transferring a texture onto a moving body. More important, this foundational research points to greater understanding of scenes for computer vision systems.
FAIR is also working to promote the development of more useful autonomous agents, both virtual assistants or robotic systems. It has has collaborated with researchers at Georgia Tech to develop a new, multistage AI task, called EmbodiedQA, that pushes the limits of reinforcement learning and natural language understanding. The team created virtual agents that must learn how to navigate computer-generated indoor spaces and how to understand and use natural language in order to answer questions about their environment. FAIR has also built a collection of virtual environments that allows agents to train dramatically faster than a physical robot would in physical spaces. These have already been open-sourced as House3D, and the data and model related to EmbodiedQA will be released in the near future:
to help the entire AI research community to make faster progress toward creating smarter, more genuinely autonomous intelligent assistants.
All in all Facebook seems to be doing a good job of democratizing AI and giving us all useful tools to use.
or email your comment to: email@example.com
|Last Updated ( Friday, 04 May 2018 )|