Microsoft Corp. has revamped one of its most popular artificial intelligence frameworks with the launch of Cognitive Toolkit 2.0 Thursday.

The product is a revamped version of Microsoft’s original packaged deep learning software CNTK, which is designed to help developers build their own neural networks. CNTK was released in beta last October, and rivals frameworks such as Google Inc.’s TensorFlow, DeepMind’s Sonnet, Facebook Inc.’s Caffe2 and Amazon Web Services Inc.’s MXNet.

Cognitive Toolkit 2.0 comes with expanded support for the Python programming language, and also for Keras, which is a popular neural network application programming interface. The new support for Keras in particular is a smart move by Microsoft. Keras is considered to be a high-level framework that’s used by lots of researchers to experiment and build machine learning models rapidly.

Microsoft also touted Cognitive Toolkit 2.0’s speech recognition credentials. Last year, Microsoft’s chief speech scientist, Xuedong Huang, announced the company had created a system that can transcribe speech at a “human parity” level, and Microsoft said the new Cognitive Toolkit is ideal for creating recurrent neural networks and long-term memory networks that are used in natural language processing. Microsoft also said the toolkit makes it easier for developers to branch out to alternative networks using Keras.

Cognitive Toolkit 2.0 also comes with Java language bindings, and new functionality for compressing trained models so they can run applications such as image recognition using less resources.

Microsoft said the framework is already being used by its digital assistant software Cortana, and that one of its goals is to make AI more accessible to developers. Microsoft Cognitive Toolkit 2.0 is available under an open-source license, and can be downloaded here.

Image: Microsoft