Last Thursday the technology announced the Open Neural Network Exchange (ONNX) format, an open source project created with Facebook, which provides a shared model of representation to ensure interoperability and innovation in the ecosystem of the intelligence framework artificial.

Microsoft’s commitment is to make artificial intelligence more accessible and valuable to all. In order to facilitate this, they offer a variety of platforms and tools, including their Cognitive Toolkit – an open source framework for building deep neural networks – and also work with other organizations that share their opinions to help the AI ​​community.

In this line, last Thursday the technological announced the Open Neural Network Exchange (ONNX) format, an open source project created together with Facebook, which provides a shared representation model to guarantee interoperability and innovation in the framework ecosystem. work of artificial intelligence. Cognitive Toolkit, Caffe2 and PyTorch technologies will support ONNX.

Thanks to the interoperability that ONNX provides, developers can move more easily between  frameworks and use the best tool for the tasks they do at hand. Each framework is optimized for specific features such as rapid training, support for flexible network architectures, inference on mobile devices, etc. Often the characteristic that is most important during research and development differs from that which is for production. This produces significant inefficiencies or delays if the appropriate framework is not used . The frameworks that ONNX uses simplify these aspects and allow developers to be more agile.

As for shared optimization (another of the advantages of ONNX) from Microsoft explain that hardware manufacturers and other companies that offer optimizations to improve the performance of neural networks can impact multiple structures at the same time if they use ONNX. Often, optimizations must be integrated separately into each framework, which can be time-consuming. Representing ONNX makes it easier for optimizations to reach more developers.