Developer

Facebook, Microsoft make AI frameworks 'interchangeable' with Open Neural Network Exchange

The two tech giants partnered on a new exchange format that aims to boost interoperability and innovation in artificial intelligence (AI) research.

A new joint effort from Microsoft and Facebook could make it easier for developers to move deep learning models between artificial intelligence (AI) frameworks. The Open Neural Network Exchange (ONNX) format, announced in blog posts from Facebook and Microsoft, looks to boost AI interoperability and innovation.

In Facebook's blog post, the social media giant describes the new effort as a step "toward an open ecosystem where AI developers can easily move between state-of-the-art tools and choose the combination that is best for them."

Often in the process of developing AI tools and technologies, developers, engineers, and researchers have to choose a specific framework based on its unique set of capabilities, the Facebook post said. However, this means that they often get locked into that framework throughout the process.

SEE: Special report: How to implement AI and machine learning (free PDF)

A big problem that's created is when the research and development phase calls for different features and capabilities than shipping to production. This had led to some inefficient workarounds, such as translating models by hand to fit different frameworks as needed, the Facebook post said.

The goal of ONNX, though, is to bridge the gap between frameworks and allow the developers and engineers to choose whichever framework best suits their project at different stages along the way. ONNX support will be coming to Caffe2, PyTorch, and Cognitive Toolkit sometime in September 2017, the Facebook post said, meaning that "models trained in one of these frameworks to be exported to another for inference."

Another feature known as shared optimization could also help advance AI development. Oftentimes, vendors and third-parties must integrate optimizations for specific AI frameworks separately, but ONNX could allow them to target multiple frameworks at once, Microsoft's post noted.

ONNX was released as an open source project, as Microsoft noted in its post, meaning that the format could see additional development and advancement from the greater open source community.

"Enabling interoperability between different frameworks and streamlining the path from research to production will help increase the speed of innovation in the AI community," the Facebook post said.

The 3 big takeaways for TechRepublic readers

  1. Facebook and Microsoft just released the Open Neural Network Exchange (ONNX), a new format allowing AI developers to more easily switch between AI frameworks.
  2. Caffe2, PyTorch, and Cognitive Toolkit will all get ONNX support in September, and the format was also released as an open source project.
  3. ONNX also allows for shared optimizations, which lets hardware vendors and other third parties to target multiple frameworks for integrations, instead of just one at a time.

Also see

ai.jpg
Image: iStockphoto/Zapp2Photo

About Conner Forrest

Conner Forrest is a Senior Editor for TechRepublic. He covers enterprise technology and is interested in the convergence of tech and culture.

Editor's Picks

Free Newsletters, In your Inbox