Microsoft Cognitive Toolkit 2.0: The Big Software Companies Behind Popular Deep Learning Frameworks
A few days ago, Microsoft announced the released of the second version of its Cognitive Toolkit. Previously known as CNTK, the Microsoft Cognitive Toolkit is an open source, deep learning framework that enables the implementation of advanced neural network models. The Cognitive Toolkit is Microsoft’s alternative to popular open source deep learning frameworks such as Theano, Torch, TensorFlow and others.
Among the notable features of the Cognitive Toolkit 2.0 we have the interoperability with Keras, a popular Python-based library that provides a high level programming model for the implementation of neural networks. Keras programs can run on deep learning runtimes such as Theano, TensorFlow and now the Microsoft Cognitive Toolkit. In its new release, the Microsoft Cognitive Toolkit also includes Java language bindings (in addition to C# and Python which were supported before) as well as a set of new tools that enable the optimization of models to run on smartphones and IOT devices. Finally, the Cognitive Toolkit also released some impressive benchmarks that show it outperforming its competitors by up to 3x on some specific scenarios.
The Microsoft Cognitive Toolkit joins the highly fragmented market of open source deep learning frameworks. In this ecosystem, many of the top software companies in the world have aligned their artificial intelligence(AI) efforts with different deep learning frameworks which makes up a fascinating competitive landscape.
Big Tech Loves Deep Learning Frameworks
Many of the top software companies in the world such as Facebook, Google, Amazon, Microsoft, Baidu or Alibaba have been regularly releasing highly sophisticated AI APIs. What is not so well-known is the fact that these same vendors have been actively supporting and pushing different open source deep learning libraries which is creating a unique combination of innovation and competition in the deep learning market. Let’s take a look:
— Google & TensorFlow: TensorFlow is arguably the best example of the link between a big tech company and an open source deep learning framework. Google open sourced TensorFlow a couple of years ago and it has made it the cornerstone of its AI strategy. In just a few months, TensorFlow has become the platform behind some of Google’s premier AI technologies such as Google ML, DeepMind Lab-Sonnet, Google AI APIs, TensorFlow Lite, etc.
— Facebook & Caffe2: Facebook has worked on several deep learning stacks like Torch. However, recently the social media giant seems to be lining up its efforts behind Caffe2 releasing several contributions such as Caffe2Go.
— Amazon & MxNet. AWS has been trailing Google Cloud and Azure on its AI efforts but recently it seems to be putting more effort behind Apache MxNet. Not as popular as some of its competitors, MxNet provides C++ and Python based programming models for the implementation of neural networks.
— Baidu & PaddlePaddle: The Google of China has also been active in the deep learning space. Baidu recently open sourced its PaddlePaddle stack which provides a Python-based programming model for the implementation of deep learning models.
— Alibaba is Playing Neutral: In a market in which cloud incumbents are actively pushing specific deep learning frameworks, Alibaba is playing a more neutral strategy. Currently, the Aliabab cloud Machine Learning service supports programs written on a variety of deep learning frameworks which makes it a very unique offering in the market.
No single deep learning framework is good at everything. TensorFlow notably excels on natural language processing and image e analysis models while the Microsoft Cognitive Toolkit has shown an impressive performance in areas such as speech recognition. With the support and resources of big tech companies, open source deep learning frameworks are likely to continue expanding its capabilities on specific areas creating a fascinating, yet very fragmented, competitive landscape.
