A new neural network platform’s been launched by Facebook, Microsoft, and Amazon that allows the interoperation of deep learning frameworks. It will enable developers to have more freedom to mix and match different models by making artificial intelligence (AI) and machine learning more accessible.
The AI platform is called the Open Neural Network Exchange (ONNX) and is basically an open-source platform that allows the combining of various AI applications into one useable application.
This makes it much easier for developers to start new projects as well as working on existing ones.
No longer will they need to keep adjusting their models to suit each individual framework they come across. And with the likes Facebook, Microsoft, and Amazon behind it, the platform is sure to be a huge success.
“ONNX is an open-source model representation for interoperability and innovation in the AI ecosystem that Microsoft co-developed,” said Microsoft as part of 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.”
Most of the top deep learning frameworks are compatible with ONNX including Microsoft Cognitive Framework, Apache MXNet, PyTorch, Caffe2, and NVIDIA TensorRT. Both Amazon and Microsoft will offer ONNX’s service through their cloud platforms.
The platform’s also compatible with on-device AI from many of the top major silicon manufacturers. This includes Intel, AMD, IBM, Huawei, ARM, Qualcomm, and NVIDIA. Being able to work with these AI technologies means deep learning models created via ONNX can be used via a supercomputer, cloud, or next-generation smartphone.
For the moment ONNX is concentrating on computer vision applications, but its operations will expand in the future.
Facebook, Amazon, and Microsoft are continually getting involved in the developer community in an attempt to develop new connections and models for the platform. Eventually, ONNX will transgress into areas such as natural language reconstruction, that require more dynamic models. For anyone looking to get more details on this to check out the company’s website.