Microsoft Cognitive Toolking Can Give TensorFlow Some Headaches
Last year, Microsoft announced the release of an open source framework called Computational Network Toolkit(CNTK to enable the implementation of deep learning capabilities. Last week, CNTK got an update and a new name: Microsoft Cognitive Toolkit.
The new version [2.0] expands the scope of its learning capabilities and the support for Python 3. Previous versions of the Cognitive Toolkit were solely based on C++ which affected its adoption by the developer community.
Cognitive Toolkit 2.0 adds other improvements including support for Visual Studio and the adoption of reinforcement learning algorithms. The latter is a particularly relevant event as reinforcement learning is becoming one of the most popular deep learning methods for training machine learning models. Microsoft also announced that new versions of the Cognitive Toolkit will include support for R and C#.
Microsoft Cognitive Toolkit is entering the crowded and yet emergent deep learning platform space that includes frameworks such as TensorFlow, Caffe, Torch, Paddle and a few others. Google’s TensorFlow is a stack that has made lot of progress recently with the release of the Google Cloud Machine Learning as well as the implementation of several mission critical solutions by Alphabet’s subsidiary: DepeMind. With this new release Microsoft Cognitive Toolkit can become a serious contender in the market. Specifically, Microsoft Cognitive Toolkit can become relevant t enterprises entering the machine learning and deep learning space. However, in order to achieve that, Microsoft still has some work to do with the Cognitive Toolkit stack
Some Ideas That Can Uniquely Differentiate Microsoft’s Cognitive Toolkit
Becoming a relevant deep learning platform in the enterprise requires more than a robust technology stack. However, Microsoft’s Cognitive Toolkit is on a very unique position to differentiate from some of the other leaders in the space. Let’s explore a few ideas that can help Microsoft’s Cognitive Toolkit accomplish that goal:
— Integration with Microsoft’s Cognitive Services: Integrating Microsoft’s Cognitive Toolkit and Microsoft’s Cognitive Services can provide a strong model for enterprises building deep learning and artificial intelligence solutions.
— Integration with Azure ML: Similarly to the model followed by Google Cloud ML, Microsoft should enable the execution of Cognitive Toolkit programs as part of Azure ML. This model will provide strong cloud and on-premise symmetry for Microsoft’s deep learning solutions.
— Integration with PoweBI: Integrating Microsoft’s Cognitive Toolkit with PowerBI will enable rich visualizations to enrich the output of Cognitive Toolkit algorithms.
— Server-Side Infrastructure and Tooling: In order to be more competitive with frameworks such as torch or TensorFlow, Microsoft’s Cognitive Toolkit should enable a more robust server side infrastructure and tooling to execute and manage programs created with the framework.
— C# and R Support: The support for R and C# can drastically increase the adoption of the Microsoft Cognitive Toolkit as well as attract a large developer community. This capability seems to already be on the roadmap of the framework