At its annual re:Invent conference, AWS today rolled out a slew of new features for SageMaker, the company’s managed service for building, training and deploying machine learning (ML) models. Swami Sivasubramanian, the vice president of machine learning at Amazon, said the new features aim to make it easier for users to scale machine learning in their organizations.
Firstly, AWS launched a new SageMaker Ground Truth Plus service that uses an expert workforce to deliver high-quality training datasets faster. SageMaker Ground Truth Plus uses a labeling workflow including machine learning techniques for active learning, pre-labeling and machine validation. The company says the new service reduces costs by up to 40% and doesn’t require users to have deep machine learning expertise. The service enables users to create training datasets without having to build labeling applications. It also allows you to manage the labeling workforce on your own. SageMaker Ground Truth Plus is currently available in Northern Virginia.
The company also rolled out a new SageMaker Inference Recommender tool to help users choose the best available compute instance to deploy machine learning models for optimal performance and cost. AWS says the tool automatically selects the right compute instance type, instance count, container parameters and model optimizations. Amazon SageMaker Inference Recommender is generally available in all regions where SageMaker is available except the AWS China regions.
In addition, AWS released the preview of a new SageMaker Serverless Interface option that allows users to easily deploy machine learning models for inference without having to configure or manage the underlying infrastructure. The new option is available in Northern Virginia, Ohio, Oregon, Ireland, Tokyo and Sydney.
With SageMaker Training Compiler, AWS today launched a new feature that can accelerate the training of deep learning models by up to 50% through more efficient use of GPU instances. The feature covers deep learning models from their high-level language representation to hardware-optimized instructions. The new feature is generally available in Northern Virginia, Ohio, Oregon and Ireland.
Lastly, AWS announced that users can now monitor and debug their Apache Spark jobs running on Amazon Elastic MapReduce (EMR) right from SageMaker Studio notebooks with just a click. The company notes that you can now also discover, connect to, create, terminate and manage EMR clusters directly from SageMaker Studio.
“The built-in integration with EMR therefore enables you to do interactive data preparation and machine learning at peta-byte scale right within the single universal SageMaker Studio notebook,” AWS explains in a blog post.
The new SageMaker Studio features are available in Northern Virginia, Ohio, Northern California, Oregon, central Canada, Frankfurt, Ireland, Stockholm, Paris, London, Mumbai, Seoul, Singapore, Sydney, Tokyo and Sao Paolo.
On a related note, AWS also today launched SageMaker Studio Lab, a free service to help developers learn machine learning techniques and experiment with the technology. Yesterday, AWS announced a new machine learning service called Amazon SageMaker Canvas. The new service will allow users to build machine learning prediction models, using a point-and-click interface.