Microsoft makes deep-learning toolkit opensource

Jan

26

2016

To the development of artificial intelligence to accelerate Microsoft’s Computational Network Toolkit, the department that conducts research into natural language institute CNTK made the open source toolkit. According to the makers institute CNTK is much more efficient than other four machine learning -toolkits

Microsoft made the toolkit already public, but under a much stricter license. Institute CNTK is under the MIT license on GitHub provided whereby others have more freedoms. The further advantage of the open source toolkit comes from the successes that have been since the kit in April last year for academic research released was booked.

According to the head of the voice department Xuedong Huang is the toolkit many times faster than comparable machine learning toolkit. The comparison is made with Theano, TensorFlow, Torch 7 and Caffe. Google made ​​TensorFlow recently been opensource.

institute CNTK microsoft speech machine learning Source: Microsoft Blog

Also institute CNTK would be the only toolkit that is scalable across multiple machines. In the test in which institute CNTK nearly twice as fast as its own program with four GPUs in a single machine, scales across two machines, each with four GPUs. According to the researchers, the program on large clusters can be scaled with many GPUs.

The program runs on Windows and Linux and is written in C ++ and is working with Nvidia’s cuDNN 4-library, but can run on CPUs, as stated in a presentation to institute CNTK late last year even indicating that the toolkit in the foreseeable future to GitHub would move.

Viewing:-147

In: Technology & Gadgets Asked By: [15174 Red Star Level]

Answer this Question

You must be Logged In to post an Answer.

Not a member yet? Sign Up Now »