MICROSOFT COGNITIVE TOOLKIT(CNTK)

Microsoft toolkit, previously know as CNTK, is a deep learning library developed by Microsoft. According to Microsoft, the library is among the fastest on the market. Microsoft toolkit is an open-source library, although Microsoft is using it extensively for its product like Skype, Cortana, Bing, and Xbox. The toolkit is available both in Python and C++.

MXNet

MXnet is a recent deep learning library. It is accessible with multiple programming languages including C++, Julia, Python and R. MXNet can be configured to work on both CPU and GPU. MXNet includes state-of-the-art deep learning architecture such as Convolutional Neural Network and Long Short-Term Memory. MXNet is build to work in harmony with dynamic cloud infrastructure. The main user of MXNet is Amazon

Caffe

Caffe is a library built by Yangqing Jia when he was a PhD student at Berkeley. Caffe is written in C++ and can perform computation on both CPU and GPU. The primary uses of Caffe is Convolutional Neural Network. Although, In 2017, Facebook extended Caffe with more deep learning architecture, including Recurrent Neural Network. caffe is used by academics and startups but also some large companies like Yahoo!.

TensorFlow

TensorFlow is Google open source project. TensorFlow is the most famous deep learning library these days. It was released to the public in late 2015

TensorFlow is developed in C++ and has convenient Python API, although C++ APIs are also available. Prominent companies like Airbus, Google, IBM and so on are using TensorFlow to produce deep learning algorithms.

TenserFlow Vs Theano Vs Torch Vs Keras Vs infer.net Vs CNTK Vs MXNet Vs Caffe: Key Differences

Library	Platform	Written in	Cuda support	Parallel Execution	Has trained models	RNN	CNN
Torch	Linux, MacOS, Windows	Lua	Yes	Yes	Yes	Yes	Yes
Infer.Net	Linux, MacOS, Windows	Visual Studio	No	No	No	No	No
Keras	Linux, MacOS, Windows	Python	Yes	Yes	Yes	Yes	Yes
Theano	Cross-platform	Python	Yes	Yes	Yes	Yes	Yes
TensorFlow	Linux, MacOS, Windows, Android	C++, Python, CUDA	Yes	Yes	Yes	Yes	Yes
MICROSOFT COGNITIVE TOOLKIT	Linux, Windows, Mac with Docker	C++	Yes	Yes	Yes	Yes	Yes
Caffe	Linux, MacOS, Windows	C++	Yes	Yes	Yes	Yes	Yes
MXNet	Linux, Windows, MacOs, Android, iOS, Javascript	C++	Yes	Yes	Yes	Yes	Yes

Verdict:

TensorFlow is the best library of all because it is built to be accessible for everyone. Tensorflow library incorporates different API to built at scale deep learning architecture like CNN or RNN. TensorFlow is based on graph computation, it allows the developer to visualize the construction of the neural network with Tensorboad. This tool is helpful to debug the program. Finally, Tensorflow is built to be deployed at scale. It runs on CPU and GPU.

Tensorflow attracts the largest popularity on GitHub compare to the other deep learning framework.

Comparing Machine Learning as a Service

Following are 4 popular DL as a service providers

Google Cloud ML

Google provides for developer pre-trained model available in Cloud AutoML. This solution exists for a developer without a strong background in machine learning. Developers can use state-of-the-art Google's pre-trained model on their data. It allows any developers to train and evaluate any model in just a few minutes.

Google currently provides a REST API for computer vision, speech recognition, translation, and NLP.

Using Google Cloud, you can train a machine learning framework build on TensorFlow, Scikit-learn, XGBoost or Keras. Google Cloud machine learning will train the models across its cloud.

The advantage to use Google cloud computing is the simplicity to deploy machine learning into production. There is no need to set up Docker container. Besides, the cloud takes care of the infrastructure. It knows how to allocate resources with CPUs, GPUs, and TPUs. It makes the training faster with paralleled computation.

AWS SageMaker

A major competitor to Google Cloud is Amazon cloud, AWS. Amazon has developed Amazon SageMaker to allow data scientists and developers to build, train and bring into production any machine learning models.

SageMaker is available in a Jupyter Notebook and includes the most used machine learning library, TensorFlow, MXNet, Scikit-learn amongst others. Programs written with SageMaker are automatically run in the Docker containers. Amazon handles the resource allocation to optimize the training and deployment.

Amazon provides API to the developers in order to add intelligence to their applications. In some occasion, there is no need to reinventing-the-wheel by building from scratch new models while there are powerful pre-trained models in the cloud. Amazon provides API services for computer vision, conversational chatbots and language services:

The three major available API are:

Amazon Rekognition: provides image and video recognition to an app
Amazon Comprehend: Perform text mining and neural language processing to, for instance, automatize the process of checking the legality of financial document
Amazon Lex: Add chatbot to an app

Azure Machine Learning Studio

Probably one of the friendliest approaches to machine learning is Azure Machine Learning Studio. The significant advantage of this solution is that no prior programming knowledge is required.

Microsoft Azure Machine Learning Studio is a drag-and-drop collaborative tool to create, train, evaluate and deploy machine learning solution. The model can be efficiently deployed as web services and used in several apps like Excel.

Azure Machine learning interface is interactive, allowing the user to build a model just by dragging and dropping elements quickly.

When the model is ready, the developer can save it and push it to Azure Gallery or Azure Marketplace.

Azure Machine learning can be integrated into R or Python their custom built-in package.

IBM Watson ML

Watson studio can simplify the data projects with a streamlined process that allows extracting value and insights from the data to help the business to get smarter and faster. Watson studio delivers an easy-to-use collaborative data science and machine learning environment for building and training models, preparing and analyzing data, and sharing insights all in one place. Watson Studio is easy to use with a drag-and-drop code.

Watson studio supports some of the most popular frameworks like Tensorflow, Keras, Pytorch, Caffe and can deploy a deep learning algorithm on to the latest GPUs from Nvidia to help accelerate modeling.

Verdict:

In our point of view, Google cloud solution is the one that is the most recommended. Google cloud solution provides lower prices the AWS by at least 30% for data storage and machine learning solution. Google is doing an excellent job to democratize AI. It has developed an open source language, TensorFlow, optimized data warehouse connection, provides tremendous tools from data visualization, data analysis to machine learning. Besides, Google Console is ergonomic and much more comprehensive than AWS or Windows.