The open source Microsoft Cognitive Toolkit 2.0 is now available for use in deep learning and artificial intelligence projects as the company provides new tools for developers as they continue to delve into new areas of research.
Microsoft announced the release of the latest toolkit in a June 1 post on the Microsoft Blog, unveiling the application almost eight months since its October 2016 beta release. The Microsoft Cognitive Toolkit (previously known as CNTK) was originally developed by the company for use by its own researchers and developers to use in deep neural network research and for applications such as video search on Bing and for a speech recognition system that can recognize the words in a conversation as well as a human.
The toolkit, which the company hopes will inspire new breakthroughs in artificial intelligence, is now available under an open source license for anyone to use, according to the company.
One of the latest features for the V2.0 release is support for Keras, which is a user-friendly open-source neural network library used by developers working on deep learning applications, according to Microsoft. Code written for Keras can now take advantage of the performance and speed available from the Cognitive Toolkit without requiring any code change. The Toolkit support for Keras is now in public preview.
In addition, the Cognitive Toolkit also supports the latest versions of NVIDIA's Deep Learning SDK and advanced graphical processing unit (GPU) architectures such as NVIDIA Volta, according to Microsoft.
The latest release version is designed for production-grade and enterprise-grade deep learning workloads and includes hundreds of new features incorporated since the beta to streamline the process of deep learning and to ensure the toolkit's seamless integration throughout the wider AI ecosystem, the company states.
A wide range of companies and organizations around the world have been using the beta release of the Cognitive Toolkit since it was released to define and train neural networks. A neural network is a system that can learn how to perform specific tasks in a way that resembles how scientists think the human brain learns, according to Microsoft.
One use of the toolkit comes from the Annapolis, Maryland-based Chesapeake Conservancy, which is working with Microsoft researchers to use the toolkit to define and train a neural network that accelerates the creation of up-to-date one-meter resolution land cover datasets. Those datasets, which are used to prioritize restoration and protection initiatives throughout the Chesapeake Bay, contain 900 times the information of existing 30-meter resolution datasets and are created in far less time than it takes to create existing datasets. Instead of requiring months of data entry and image processing, the new neural network compresses the workflow into a single algorithm that can produce a map in a fraction of the time. The AI technology could scale to aid national and global conservation efforts, according to project partners.
"Originally, people handwrote their own mathematical functions and created their own neural networks with their own private code and figured out how to feed it with data all by themselves," Chris Basoglu, a partner engineering manager at Microsoft, said in a statement. "But now the data is so large, the algorithms are so complex and optimization across multiple GPUs, CPUs and machines is so prohibitive that it is not feasible for someone to write their own. They need tools."
Basoglu and his team built the tools so they can be used by both highly-skilled developers to train their own deep neural networks with massive datasets across multiple servers running the latest GPUs or by enthusiasts with basic programming skills and a laptop.
Several IT analysts told ITPro that the latest toolkit is a helpful addition for researchers.
"Coming just seven months after the release of the beta version, Microsoft's Cognitive Toolkit 2.0 qualifies as a reaffirmation of the company's commitment to fostering deep learning and other artificial intelligence (AI) development, processes and projects," Charles King, principal analyst with research firm Pund-IT, said in an email reply to an inquiry. "There are good reasons for Microsoft's enthusiasm, including providing support for AI efforts via its Azure public cloud."
In addition, the company's release of the toolkit as an open source offering is intriguing, especially since former company CEO Steve Ballmer once called Linux "a cancer," wrote King. "Microsoft's current CEO Satya Nadella has a far more genial view of open source that is informed by his long experience and deep expertise in software development."
Ultimately, the company's embrace of open source in AI is "critical for Microsoft and every any other vendor hoping to develop a business around AI," he wrote. "Virtually all of the mainstream efforts related to AI are leveraging open source assets and tools, and the developers involved have shown a clear preference for open source methodologies. The release of the new Cognitive Toolkit 2.0 demonstrates that Microsoft clearly understands these market dynamics and is committed to playing by the rules."
Another analyst, Dan Olds, principal of Gabriel Consulting Group, agreed.
"There's a big battle, which will be soon be bigger, among industry players about what will be the platform of choice for Deep Learning, Machine Learning, and AI," said Olds. "This is what's behind Microsoft's move to open source their Cognitive Toolkit. They want to attract users to their platform and, hopefully, turn them into steady customers. They'll face tough competition from everyone else and their own AI frameworks, many of which are already open source."
Microsoft could have an advantage, though, because "most users probably are already familiar with its products, which could give Microsoft a leg up in this case."