Framework Optimizations
Featured CPU-optimized deep learning topologies for several popular frameworks – check back for regular updates.
| AI Capability | Application | Tensorflow | MXNet | Caffe2 | Caffe |
|---|---|---|---|---|---|
| Image Recognition | Classify image(s) with high accuracy | ResNet50 (I&T) InceptionV3 (I&T) |
ResNet50 (I&T) InceptionV3 (I&T) InceptionV4 (I&T) MobileNet (I) SqueezeNets (I) DenseNet (I) |
ResNet50,101(I) InceptionV1 (I) SqueezeNet (I) VGG16,19 (I) |
ResNet50 (I&T) InceptionV3 (I&T) |
| Object Detection & Localization | Locate and classify object(s) in image | SSD-VGG16 (I&T) | SSD (I) | SSD-VGG16 (I&T) Faster-RCNN (I&T) R-FCN (I&T) |
|
| Speech Recognition | Convert speech to text | Deep Speech 2 (I) | |||
| Language Translation | Translation from one language to another | GNMT (I) | NMT (I) Transformer (I) |
||
| Recommender Systems | Predicts the rating or preference a user would give an item | Wide & Deep (I) | |||
| Generative Adversarial Networks | Neural networks that generate data mimicking some distribution | DCGAN (I) | |||
| Reinforcement Learning | The use of actions and results to learn how to behave in an environment | A3C (I) |
This Python*-based deep learning framework is designed for ease of use and extensibility on modern deep neural networks and has been optimized for use on Intel® Xeon® processors.
The open-source, deep learning framework MXNet* includes built-in support for the Intel® Math Kernel Library (Intel® MKL) and optimizations for Intel® Advanced Vector Extensions 2 (Intel® AVX2) and Intel® Advanced Vector Extension 512 (Intel® AVX-512) instructions.
Based on Python* and optimized for Intel® architecture, Intel’s innovative neon™ framework for deep learning is designed for ease of use and extensibility on modern deep neural networks.
The Intel® Optimization for Caffe* provides improved performance for of the most popular frameworks when running on Intel® Xeon® processors.
Theano*, a numerical computation library for Python, has been optimized for Intel® architecture and enables Intel® Math Kernel Library (Intel® MKL) functions.
Chainer* is a Python*-based deep learning framework for deep neural networks. Intel’s optimization for Chainer is integrated with the latest release of Intel® MKL-DNN.