Deep Learning with TensorFlow
Deep learning on AWS made simple
TensorFlow is one of many deep learning frameworks available to researchers and developers to enhance their applications with machine learning. AWS provides broad support for TensorFlow, enabling customers to develop and serve their own models across computer vision, natural language processing, speech translation, and more.
You can get started with TensorFlow on AWS using Amazon SageMaker, a fully managed machine learning service that makes it easy and cost-effective to build, train, and deploy TensorFlow models at scale. If you prefer to manage the infrastructure yourself, you can use the AWS Deep Learning AMIs or the AWS Deep Learning Containers, which come built from source and optimized for performance with the latest version of TensorFlow to quickly deploy custom machine learning environments.
TensorFlow comes with a full suite of visualization tools that make it easy to understand, debug, and optimize applications. With support for a variety of styles – from images and audio to histograms and graphs – you can train massive deep neural networks quickly and easily.
With TensorFlow, you get access to extensive documentation and tutorials that can help accelerate your AI development. TensorFlow also has a large and extremely active community of users who regularly contribute code and resolve issues on GitHub.
Aerobotics, a South African agri-tech startup, provides farmers with data and intelligence through the Aeroview platform, which utilizes machine learning to extract information from aerial drone images. Using Amazon SageMaker and TensorFlow, Aerobotics was able to improve their training speed by 24 times per sample.
Fannie Mae uses Amazon SageMaker with TensorFlow for its home appraisal model to generate accurate property valuations, which helps reduce mortgage risk.
Mobileye, an Intel company, uses TensorFlow with Amazon SageMaker to develop and deliver driver assistance and autonomous vehicle solutions. Using Amazon SageMake Pipe Mode, data is streamed from Amazon S3 to training instances with the TensorFlow dataset API to allow multiple training instances to pull from the same set of decoupled training data.
Blogs and articles
Blog: Building a customized recommender system in Amazon SageMaker
by Ray Li
Blog: Deploying Tensorflow Openpose on AWS Inferentia based Amazon EC2 Inf1 instances
by Fabio Nonato de Paula and Haichen Li
Blog: Train ALBERT for natural language processing with TensorFlow on Amazon SageMaker
by Jared Nielsen, Aditya Bindal, and Derya Cavdar
Blog: A quick guide to distributed training with TensorFlow and Horovod on Amazon SageMaker
by Shashank Prasanna