Microsoft/CNTKArchived

Features

• Deep Learning Toolkits - Functions as a deep learning toolkit for designing, constructing, and training complex neural networks.
• Model Construction - Provides a comprehensive toolkit for the construction and training of neural network architectures.
• Automatic Differentiation Engines - Provides an automatic differentiation engine to compute gradients for network parameter optimization.
• Computational Graph Definitions - Allows the definition of deep learning models as directed graphs of computational steps.
• Computational Graphs - Defines neural network architectures as directed graphs of operations to manage data flow.
• Distributed Training - Spreads training workloads across multiple hardware nodes and GPUs to handle large scale models.
• Data-Parallel Training - Distributes training workloads across multiple hardware nodes and GPUs to accelerate large dataset processing.
• Large-Scale Model Training - Enables training of large-scale models by distributing workloads across multiple GPUs and hardware nodes.
• Neural Network Architectures - Provides tools for defining neural network architectures as computational graphs with specialized layers.
• Neural Network Parallelism - Employs data and model parallelism to distribute computational work across multiple graphics processors.
• Neural Networks - Implements foundational computational models and training processes for complex pattern recognition.
• Stochastic Gradient Descent - Uses stochastic gradient descent and automatic differentiation to optimize network parameters.
• Dynamic Sequence Convolutions - Performs convolutions across both spatial and dynamic sequence axes to process variable-length data.
• CPU Optimizations - Optimizes tensor operations and floating-point math on CPUs to increase model prediction speeds.
• Inference Accelerations - Offers optimized floating-point math and tensor kernels to accelerate model predictions on Intel processors.
• Hardware Acceleration Kernels - Employs optimized tensor kernels and GPU instructions to speed up linear algebra and convolutional operations.
• Convolution Layers - Implements convolutional layers with padding algorithms to ensure symmetric boundaries and performance.
• Grouped - Implements hardware-accelerated grouped and depth-wise convolutions to reduce memory and execution time.
• ONNX Model Exports - Supports the exchange of neural network models via the Open Neural Network Exchange format.
• Model Interoperability Formats - Supports model interoperability by converting internal network structures into the standardized ONNX format.
• Model Interoperability Tools - Facilitates compatibility across different deep learning toolkits using the ONNX specification.
• Dynamic Sequence Convolutions - Implements specialized convolution filters that process data of variable lengths across dynamic sequence axes.
• AI & Machine Learning - Deep-learning toolkit.
• Artificial Intelligence - Deep learning toolkit for distributed training and inference.
• Computer Vision - Handwritten digit recognition using convolutional neural networks.
• Deep Learning Frameworks - Distributed deep learning toolkit for high performance.
• General Machine Learning - Unified deep learning toolkit using directed graphs for neural networks.
• Machine Learning and AI - Open-source deep learning toolkit.
• Machine Learning Frameworks - Deep learning toolkit designed for performance and scalability.

Star history