Building a neural network is hard. Most developers working on artificial intelligence projects are forced to commit early to a framework developed by experts in the field, and if they later realize they needed to go in a slightly different direction, it’s much harder to switch. Microsoft and Facebook have decided to make that easier.
The two companies announced the creation of the Open Neural Network Exchange (ONNX) format Thursday. The format allows AI developers to switch between open-source frameworks like Facebook’s Caffe2, Microsoft’s Cognitive Toolkit, and PyTorch, the three frameworks supported at the moment.
AI frameworks give developers a road map of sorts for how their applications will learn from the training data supplied to those models. They’re powerful tools when applied to the right application, but this field is so new that a lot of developers don’t necessarily know which approach is right for their situation until they’ve gotten pretty far down the road with their projects.
ONNX will allow developers to more easily switch between different frameworks within the context of their applications, or use multiple frameworks to train different parts of their applications. Ideally, this will convince more developers to start experimenting with AI models for their applications knowing they’ve got an out should they back themselves into a corner with the framework they chose at the outset.
“Enabling interoperability between different frameworks and streamlining the path from research to production will help increase the speed of innovation in the AI community,” Facebook said in a blog post announcing the new effort. More details will be released over the course of the month.