Microsoft makes its deep learning framework accessible for developers


Barely a week seems to go by nowadays without Microsoft Corp. making some gesture of support towards the developer community. Last Tuesday, the software giant pledged to increase funding for cloud projects that promote the public good, and now it’s expanding the range of applications where its open-source deep learning framework can be put to use with a similar goal in mind.

The Computational Network Toolkit, or CNTK, was built at Redmond’s speech recognition laboratory to speed the development of the complicated algorithms needed for services such as Cortana and Skype Translator. Creator Xuedong Huang and his team gave the framework the ability to scale across a large number of machines in order to let Microsoft’s deep learning researchers simulate the massively distributed environment where the data from its different products is processed. The idea is that the more closely real-life operational conditions can be approximated during the testing of a model, the better its accuracy will turn out in production.

The functionality could provide the same benefits for web-scale deep learning projects at other companies. However, the license under which CNTK was originally released last year restricted usage to academic research. Today’s update modifies the terms to allow for commercial deployments as well in an effort to improve the quality of the speech and image recognition software available on the market. Microsoft is bringing the framework to GitHub on occasion of the change in an effort to raise awareness among developers.

The move ups the ante against the other web giants that have open-sourced their internal deep learning technologies over recent months. Google Inc. set the bandwagon in motion with the release of its internal TesnorFlow engine in November, a move that quickly prompted responses from IBM Corp., Facebook Inc. and, of course, Redmond. The most recent addition to the roster is Baidu Inc., which released the machine learning library that supports its voice search feature under a free license earlier this month.

Image via Pixabay