Microsoft has come up with first beta’s next version of CNTK, it’s called Microsoft Cognitive Toolkit, earlier known as ‘Computational Network Toolkit’. The new system is introduced to speed up some areas such as image and speech recognition. It has also been introduced to make advances in search relevance of NVIDIA GPUs and CPUs.
Microsoft Cognitive Toolkit which enables you to tackle the intelligence with the huge datasets through deep learning which has great support for scaling, speed, and accuracy without any compromise in commercial-grade quality and compatibility with the programming languages and algorithms which you are already using.
Xuedong Huang, Microsoft’s chief speech Scientist told that Microsoft Cognitive Toolkit gives you the better performance and has many advantages when compared with TensorFlow, which is also a similar framework supported by Google.
Huang also argued saying that Cognitive Toolkit maintains to perform better than its competitors in most of the tests. Undoubtedly, this latest version is much faster than the previous releases, specifically when working with big datasets. This works well for single-GPU performance, but Huang argues saying that CNTK already has the capability of utilizing a large set of GPUs and this fact implies to the latest version as well.
Microsoft can give rights for other companies for the Toolkit production since the Toolkit is open-source. The Toolkit also has its uses to develop the commercial grade AI, which includes products like Xbox, Cortana, Bing, and Skype. The latest release of the Toolkit is available on GitHub and comes with complete APIs for defining the learners, networks, and readers along with the evaluation from Python, BrainScript, and C++.
According to Allison Linn, blog post writer at Microsoft, the Toolkit has better performance when compared to all the previous versions. In order to meet the high speed and high level of intelligence needed by Microsoft, similar improvement like the current version is required for deep learning across multiple GPUs.