Microsoft Releases Updated Cognitive Toolkit

Microsoft has released a beta version of its updated Cognitive Toolkit, the deep learning technology that its developers use to train computer systems in speech and image recognition.

Available on Github under an open source license, the toolkit is the same one Microsoft researchers used to create a speech recognition system that could recognize words as well as human transcriptionists, a breakthrough the company announced last week.

The new version allows programmers to use Python and C++ languages when working with the toolkit. In a blog post published Tuesday, Microsoft engineering manager Chris Basoglu said the company first built the kit as a ready-made solution that would allow developers to experiment with deep learning without writing a lot of extra code. But developers asked for more ability to integrate their own code.

Researchers can also use the new toolkit to program reinforcement learning, a type of artificial intelligence training in which the system learns the right way to do something through trial and error.

Microsoft is touting the Cognitive Toolkit’s ability to work efficiently across multiple servers, so researchers can wrangle large datasets.

“Democratizing AI” has become the new buzzphrase in Redmond, and Microsoft appears to be working hard to rapidly develop new AI technology and share it with developers. Besides powering Skype translator and the Cortana digital assistant, the company’s deep learning system is also beginning to be deployed in industrial settings. Appliance company Liebherr is using it to teach refrigerators how to recognize items on their shelves and automatically update grocery shopping lists.