Despite the many advances in artificial intelligence (AI) and machine learning in recent years, researchers have not yet been able to develop a computer that can understand natural-language speech as well as humans. There are plenty of people working on the problem, though, which is why Microsoft is making its Computational Network Toolkit available to developers on an open source basis.
Called CNTK for short, the toolkit has already proved far more efficient than others created to build deep-learning models for machine-enabled communication, according to Microsoft.
Aiming to speed up advances in machine-based speech and image recognition, Microsoft researchers who developed the CNTK toolkit first made it available to outside academic researchers last spring through the Codeplex open source software project hosting site. Yesterday, however, they moved the toolkit to GitHub to make the code more widely available to any developers interested in advancing deep-learning capabilities.
Training Neural Networks '10 Times Faster'
Part of CNTK's effectiveness for modeling deep learning stems from its use of computers with graphics processing units, or GPUs. Originally designed to handle computer graphics, GPUs have also proved useful for handling a number of other complex computing tasks simultaneously, in particular for processing algorithms for speech, speech comprehension and image/motion recognition.
Microsoft said that CNTK has, for example, allowed it to make its Cortana digital assistant even smarter. The company has also used its advances in machine learning to enable services like real-time translation, something it has already rolled out for several languages to Skype.
"The combination of CNTK and Azure GPU Lab allows us to build and train deep neural nets for Cortana speech recognition up to 10 times faster than our previous deep learning system," chief speech scientist Xuedong Huang (pictured above) wrote last month in a Microsoft Research blog post. "We've seen firsthand the kind of performance CNTK can deliver, and we think it could make an even greater impact within the broader machine learning and AI community."
Aiming for AI Breakthroughs
One of the advantages of CNTK is that it can scale easily from one computer to many, according to a new Microsoft blog post published yesterday. That enables the toolkit to be used by AI researchers whether they're working with just one computer or with a large cluster of GPU-based machines, said principal development manager Chris Basoglu.
Microsoft is not the only tech company working to advance developments in machine learning. In November, Google also made its machine-learning system, TensorFlow, available as open source, and last week announced it was collaborating with online learning site Udacity to offer a course in deep learning.
With Microsoft adding CNTK to the open source mix this week, AI researchers will have another tool to help them improve machine-learning capabilities. Microsoft said its toolkit could also prove useful to deep-learning startups and established companies that need to process large volumes of data in real time.
"With CNTK, they can actually join us to drive artificial intelligence breakthroughs,"" Huang said.
Image Credit: Microsoft. Photo by Scott Eklund/Red Box Pictures.