The Complete Magazine on Open Source

Microsoft’s Cognitive Toolkit 2.0 debuts with open source neural library Keras

3.8K 0

Cognitive Toolkit 2.0

Microsoft’s deep learning and artificial intelligence (AI) solution Cognitive Toolkit has reached version 2.0. The new update is designed to handle production-grade and enterprise-grade deep learning workloads.

To enable the developments around deep learning applications, Cognitive Toolkit 2.0 supports Keras, an open source neural networking library. The latest integration helps developers receive a faster and more reliable deep learning platform without bringing any code changes.

Chris Basolu, a partner engineering manager at Microsoft — who has been playing a key role in developing the Cognitive Toolkit builds — has tweaked the existing tools to make them accessible to enthusiasts with basic programming skills. There are also customisations available for highly-skilled developers who are set to accelerate training for their own deep neural networks with large-size datasets and across multiple servers.

Cognitive Toolkit 2.0 supports latest versions of Nvidia Deep Learning SDK and advanced GPUs like Nvidia Volta.

Java binding with model evaluations

In addition to the Keras support, the latest Cognitive Toolkit brings model evaluations using Java bindings. There are also few other tools that compress trained models in a real time. Further, the new version is capable of compressing models even on resource-constrained devices such as smartphones and embedded hardware.

Microsoft has designed the AI solution after observing the problems faced by smart startups to large tech companies, government agencies and NGOs. Moreover, the team led by Basolu targets the major Cognitive Toolkit features to a wide range of audience.

Developers with the previous release are recommended to upgrade to the latest Microsoft Cognitive Toolkit 2.0. You can find the code of the latest version on a GitHub repository.