Amazon Web Services Inc. is throwing its weight behind an open-source tool that makes it easier for artificial intelligence developers to switch their projects to different software frameworks without having to rewrite them from scratch.

The Open Neural Network Exchange project is a format for deep learning neural networks, which enable machines to learn tasks rather than being explicitly programmed, that was first unveiled in September by its creators Microsoft Corp. and Facebook Inc. At the time, the two companies said ONNX was intended to be a standardized format that will allow deep learning models trained on one framework to be transferred to another with minimal extra work.

Microsoft and Facebook built ONNX because they said deep learning is still a relatively new field, and there aren’t many methods available for developers to build data models. More importantly, it’s difficult and time-consuming to switch deep learning models to different frameworks from the one they’re originally built on.

ONNX was therefore designed as an easy way for developers to transfer their deep learning models among several popular frameworks. They include Microsoft’s Cognitive Toolkit, Facebook’s Caffe2 and PyTorch, which is an open-source deep learning framework also backed by Facebook as well as Twitter Inc., Salesforce.com Inc., and Uber Technologies Inc., among other companies.

Now, with Amazon joining the party, developers can also bring their deep learning models to the Apache MXNet framework, the company said. The news is a positive development for those working in AI, as the general promise of ONNX is that it will make it much easier for developers to add deep learning capabilities to their applications.

However, there’s one elephant that isn’t in the room yet: Google LLC, which developed the TensorFlow deep learning framework that it has since open-sourced. On Tuesday, Google announced a lightweight version called TensorFlow Lite in a developer preview. It’s intended to allow machine learning models to run on mobile and embedded devices.

Image: ColiNOOB/Pixabay