Microsoft Cognitive Toolkit (previously CNTK)

Microsoft Cognitive Toolkit Deep Learning software enables users to train deep learning algorithms and create deep learning systems with ease.

Overview

Microsoft Cognitive Toolkit Deep Learning software is an open-source toolkit that users use for enterprise-grade distributed deep learning. Users can use this software to teach deep learning algorithms to learn like the human brain. This software uses a directed graph to describe a neural network as a series of computational steps, and users can use it to combine and create popular model types like convolutional neural networks, feed-forward DNNs, and recurrent neural networks. Plus, the Microsoft Cognitive Toolkit Deep Learning software implements stochastic gradient descent learning with automatic parallelization and differentiation across different servers and GPUs.

This software allows users to use deep learning to tackle the intelligence within large scale datasets. Microsoft Cognitive Toolkit Deep Learning software offers users uncompromised accuracy, scaling, and speed with commercial-grade compatibility and quality with algorithms and programming languages that they already use. With its framework functions written in C++, users can call its service using C++ and Python. Besides, Microsoft offers users an introductory course to deep learning with the Microsoft Cognitive Toolkit Deep Learning software.

Users can use the Microsoft Cognitive Toolkit Deep Learning software as a library in their Python, C++, or C# programs, and they can use their model description language on it as a standalone machine-learning tool. This software allows users to use its model evaluation functionality from their Java programs with ease. Microsoft Cognitive Toolkit Deep Learning software supports 64-bit Windows and Linux operating systems, and users can choose the pre-compiled binary packages or get the toolkit from the source in GitHub. Additionally, the software supports CUDA 10 for Linux and Windows systems, and it completely supports the Open Neural Network Exchange (ONNX) format.

Product Details

Microsoft Cognitive Toolkit Deep Learning software's in-built components help users to handle multi-dimensional sparse and dense data from BrainScript, C++, and Python. This software offers users Sequence-to-Sequence with attention, Batch normalization, CNN, FFN, and RNN. Users can use this software to add new user-defined components on the GPU from Python. The Microsoft Cognitive Toolkit Deep Learning software optimizes its in-built readers for large datasets. Users can carry out reinforcement learning, supervised and unsupervised learning, and generative adversarial networks seamlessly. Plus, this software provides users with automatic hyperparameter tuning.

Microsoft Cognitive Toolkit Deep Learning software offers users full APIs to define learners, networks, evaluation, training from BrainScript, C++, and Python. Users can use C#, C++, Python, and BrainScript languages to evaluate models seamlessly. This software offers user interoperation with NumPy, and automatic shape inference based on their data.  Microsoft Cognitive Toolkit Deep Learning software provides users with low-level and high-level APIs for flexibility and ease of use. Besides, this software offers users completely optimized symbolic RNN loops that do not need any unrolling.

Microsoft Cognitive Toolkit Deep Learning software allows users to train and host their deep learning models with Azure. When users use this open-source toolkit with Azure GPU and Azure networks, it takes advantage of high-speed resources. This software enables users to host trained models on Azure with ease, and they can add real-time training if required. Microsoft Cognitive Toolkit Deep Learning software offers users parallelism with accuracy on diferent GPUs using 1-bit SGD and Block Momentum. Additionally, this software can fit the most significant models in GPU memory with memory sharing and different in-built methods. 

Microsoft Cognitive Toolkit Deep Learning software scales efficiently from different environments while maintaining accuracy. Users can use this software to customize any of the in-built training algorithms or use thier own, and they can use it to measure model performance. This software offers users different components to use to measure the performance of the neural networks they build. Also, the Microsoft Cognitive Toolkit Deep Learning software generates log data from the different models to enable users to monitor the training process seamlessly.

Microsoft Cognitive Toolkit Deep Learning software supports the Open Neural Network Exchange format. The ONNX format is an open-source shared model representation used for shared optimization and framework interoperability. With the ONNX format, developers can move models between frameworks like PyTorch, MXNet, Caffe2, and CNTK. Users can use the Microsoft Cognitive Toolkit Deep Learning software to export over 2GB models in ONNX format. Plus, this software supports FP16 in the training action of BrainScript scripting language. 

Microsoft Cognitive Toolkit Deep Learning software allows users to train models using declarative and imperative API. Users can train their model using the API by calling the training method and providing the configuration parameters for various training aspects like checkpointing, data sources, progress printing, and cross-validation. With the high-level declarative API, this software eliminates boilerplate code and streamlines the implementation of recurrent training tasks. Besides, users can use the Microsoft Cognitive Toolkit Deep Learning software to specify how often they can save a checkpoint to a given file.

Microsoft Cognitive Toolkit Deep Learning software allows users to evaluate the trained model in their application. This software supports model evaluation from Python, .NET, Java, and C++, and it supports the Universal Windows Platform. Users can use this software on both CPU and GPU devices, and it supports different evaluation requests in parallel. Microsoft Cognitive Toolkit Deep Learning software shares the same model between different threads to optimize memory usage. Additionally, this process reduces memory usage when =carrying out an evaluation in a service environment significantly.

Microsoft Cognitive Toolkit Deep Learning software allows users to evaluate a model in an Azure Web API. Users can deploy a CNTK model on Azure and run the deployed model through Web APIs using a command-line interface to Azure Machine Learning. This software offers users CPU-only and GPU-only build versions, and users can install both versions with ease. Also, the Microsoft Cognitive Toolkit Deep Learning software supports the NVIDIA graphics card.

Recap

Microsoft Cognitive Toolkit Deep Learning software is an open-source toolkit that focuses on training deep learning algorithms to learn like the human brain. This software allows users to use deep learning alongside their programming algorithms and languages to harness the intelligence within massive datasets.