Microsoft is seeking to spread usage of artificial intelligence (AI) deep learning software by moving its open source Computational Network Toolkit (CNTK) on GitHub.
Since April 2015, CNTK has been available via open source but under a more-restrictive license. Now, Microsoft made it available to any developer with an MIT license. The kit provides a high performance unified computational network framework that describes deep neural networks as a series of computational steps.
According to a Microsoft blog post by Allison Linn, Microsoft Research developed out of necessity the CNTK open source toolkit with the aim to make faster improvements to computer speech recognition software. Previous tools such as Torch 7, Theano, Caffe and TensorFlow were slowing down the research. Microsoft Research internal tests come to found CNTK more efficient.
By moving its open source on GitHub, Microsoft made it now more community-focused, according to analyst Rob Sanfilippo of Microsoft. The high tech giant is striving to gain mind share for solutions that have not been adopted before due to the opacity of the software or licensing requirements. Microsoft cannot longer avoid the open source trend in software.
According to InfoWorld, Sanfilippo added that this new strategy could help Microsoft by sparking interest in other services, such as Azure. Developers working on neural networks, deep learning, image and speech recognition, will be interested in evaluating CNTK. Microsoft's toolkit can help them speed projects due to performance gains, in particular on multi-GPU hardware.
Microsoft tested CNTK on computers using GPUs and was able to achieve advances in technology that recognize movements and images, that can hear, understand and speak. Now the high tech giant plans to leverage Azure GPU Lab and CNTK in order to provide a distributed GPU platform for AI research.
According to Linn, the field of artificial intelligence and deep learning has grown in recent years. An increasing number of researchers are running now computer learning algorithms on deep neural networks in order to improve artificial intelligence.
