Email
Password
Remember meForgot password?
    Log in with Twitter

article imageMicrosoft, Amazon and Facebook launch new deep learning toolkit

By James Walker     Dec 7, 2017 in Technology
Microsoft, Amazon and Facebook have announced the release of a neural network platform that allows deep learning frameworks to interoperate. It's designed to make machine learning more accessible by letting developers mix and match different models.
First announced back in September as a Microsoft and Facebook project, Amazon Web Services has now got onboard too. Called Open Neural Network Exchange (ONNX), the platform is an "open source model representation" that makes it possible to combine different components of the AI ecosystem into a single application.
Using ONNX, developers can utilise deep learning models from different frameworks. This makes it easier to get started on a new service and improves interoperability between frameworks. Developers can focus on creating new original functionality, instead of having to adjust their models to suit each new framework they work on.
"ONNX is an open source model representation for interoperability and innovation in the AI ecosystem that Microsoft co-developed," explained Microsoft in its announcement post. "The ONNX format is the basis of an open ecosystem that makes AI more accessible and valuable to all: developers can choose the right framework for their task, framework authors can focus on innovative enhancements, and hardware vendors can streamline optimizations."
READ NEXT: Report: Consumers sceptical of AI, fearful of privacy risks
Several of the most popular deep learning frameworks are now compatible with ONNX. These include Apache MXNet, Caffe2, Microsoft Cognitive Framework, NVIDIA TensorRT and PyTorch. In addition, community submitted connectors are available for frameworks such as Google's TensorFlow. ONNX will be available on the cloud platforms offered by both Amazon and Microsoft.
The platform also works with on-device AI from all the major silicon manufacturers. AMD, ARM, IBM, Intel, Huawei, NVIDIA and Qualcomm have committed to supporting ONNX with their upcoming chips. This will allow a deep learning model created with ONNX to be deployed to the cloud, a supercomputer or a next-generation smartphone.
At launch, ONNX is focusing on computer vision applications but its scope will expand in future versions. Microsoft, Amazon and Facebook are actively involving the developer community to help create new models and framework connections for the platform. Over time, it's anticipated ONNX will move into areas that require dynamic models, such as natural language reconstruction. Developers can get more details on the platform at its dedicated website.
More about onnx, Ai, Artificial intelligence, Developers, computer vision