Microsoft, Facebook Announce Open Neural Network Exchange, Simplify PyTorch to Caffe2 Conversion

Microsoft, which last week announced it is was partnering with Amazon to let their respective AI-powered virtual assistants talk, said this week it has joined forced with Facebook to launch an open source AI resource repository called Open Neural Network Exchange (ONNX).

The AI resource repository, the companies said, will allow developers to swiftly switch between the company's respective AI engines - PyTorch and Caffe2 - at any stage of the development. The Open Neural Network Exchange addresses one of the key issues that is hindering the growth of the machine learning ecosystem. There are various frameworks for executive neural networks but they are all different and not interoperable.

Developers have long desired for a common ground among these different frameworks, as each offers its own advantages. Kovas Boguta, a developer at Twitter, said "Looks like the long-awaited 'export pytorch to caffe2' has dropped. Interesting development."

Facebook maintains two different AI modules -- FAIR and AML. The company uses FAIR to handle bleeding edge research, while AML to bring AI-powered solutions to consumer-facing services. FAIR supports PyTorch, while AML supports Caffe2. The collaboration between Facebook and Microsoft will enable developers to easily convert models built in PyTorch into Caffe2 models.

It's a welcome move from the two companies, but developers who prefer Google's TensorFlow and other key frameworks, and Apple's CoreML are still in their wallet chambers, as both only allow limited conversions to other models.

"People experimenting with new models, and particularly those in research, want maximum flexibility and expressiveness in writing neural networks - ranging from dynamic neural networks to supporting gradients of gradients, while keeping a bread-and-butter ConvNet performant," developers at Facebook wrote.

"Researchers also want to iterate rapidly, which means that they need excellent tooling for interactive development and debugging. PyTorch has been built to push the limits of research frameworks, to unlock researchers from the constraints of a platform and allow them to express their ideas easier than before."