Microsoft unveiled an update to its Cognitive Toolkit.

Yesterday, Microsoft, in its official blog, presented to the public an update of its Microsoft Cognitive Toolkit, a system for designing and training deep, convolutional and recurrent networks.

In addition, the Cognitive Toolkit makes it possible to use reinforcement training . Cognitive Toolkit is proposed to be used in such areas as object recognition, speech and search for relevant results.

The Cognitive Toolkit was based on another company's development, the Computational Network Toolkit, which is available on GitHub . It also directly indicates that CNTK received a new version and the name: "CNTK becomes The Microsoft Cognitive Toolkit".
')
The announcement of the CNTK as a tool took place in January of this year, including at Habré in the official Microsoft blog. Initially, the tool was designed as a competitor TensorFlow from Google and warp-ctc from Baidu.

The strength of the Cognitive Toolkit indicates the ability to work on the basis of a conventional PC and use both the CPU and gaming video cards with a powerful NVIDIA GPU.

“Microsoft Cognitive Toolkit is a product of a collaborative collaboration between Microsoft and NVIDIA, the goal of which was the progress of Glybine learning in the community,” the company's official blog said.

The CNTK kernel and, as a result, the Cognitive Toolkit, is implemented in C ++ with the ability to manage in C ++ and Python. The latest update (and the transition from CNTK to Cognitive Toolkit) added support for the Python API and the Fast R-CNN algorithm.

» CNTK / Cognitive Toolkit page on GitHub