Microsoft and Amazon Web Services (AWS) co-operate and jointly present the deep-learning interface Gluon. Both companies have issued an online report. Gluon will be source for both the Microsoft Cognitive Toolkit and Apache Mxnet.
Gluon is intended to be used for instructional, imperative programming and self-manipulable symbolic programming. The second programming paradigm is particularly interesting in the development of artificial intelligence. According to AWS, networks are made up of easy-to-understand nodes. Developers should get an easy introduction to the programming of neural networks.
In addition, the API is designed for dynamically extensible neural networks. A Gluon network should be able to train while it is running. This is not possible with many other networks, which must always be stopped. Developers can also directly influence a running network using standard algorithms, branches, and loops.
Scalability and handling of sparse data
Gluon is closely linked to the backends of Mxnet and the Cognitive Toolkit, thus enabling the automatic distribution of compute load on several graphics processors. Microsoft states that a training service can scale linearly to more than 500 GPUs.
The API also supports models that can deal with sparse data rather than big data. In addition, a trained model can be exported to local devices, such as smartphones or notebooks, and used there – hardware, on which usually only a small amount of data is available.
Gluon is already available for Apache Amxnet on the developer portal Github . Later, the support for Microsoft’s Cognitive Toolkit follows. The API can be used on platforms such as Azure Data Science Virtual Machine or Azure Machine Learning.
Gluon is not the first collaboration between Microsoft and another great partner. Together with Facebook , the company introduced ONNX , an open-source AI framework. KI and Machine Learning are supposed to be made available to everyone in general.