Amazon on Wednesday introduced a series of enhancements to SageMaker[1], its fully-managed machine learning service. SageMaker Workflows is a series of features that makes it easier to manage machine learning pipelines. Amazon is also introducing new built-in algorithms and new framework support, as well as new compliance standards and accreditation.
Amazon rolled out SageMaker at last year's AWS re:Invent conference[2] -- along with a slew of other new services[3], many powered by machine learning. Ahead of this year's conference, Amazon Web Services has announced several improvements to those services, including updates[4] to Amazon Polly, Transcribe and Translate. Amazon also just introduced predictive scaling[5] for EC2 instances. SageMaker itself has seen nearly 100 new features added in the past year, Amazon noted.
With SageMaker Workflows, customers are getting new automation, orchestration, and collaboration features for machine learning pipelines. For instance, SageMaker Search lets customers quickly find relevant model training runs, right from the AWS console. This will help them more easily find the right combination of datasets, algorithms and parameters for their models.
Workflows also includes Git integration and visualization for better collaboration and version control. Additionally, customers can now use Step Functions to automate and orchestrate SageMaker steps in an end-to-end workflow. SageMaker also now integrates with Apache Airflow, a popular open source framework, to author, schedule and monitor multi-stage workflows.
Amazon is also introducing new algorithms to SageMaker for detecting suspicious IP address (IP Insights), low dimensional embeddings for high dimensional objects (Object2Vec), and unsupervised grouping (K-means clustering). These built-in algorithms are all designed to support petabyte scale datasets. Amazon has also been adding new framework support. Soon, customers will be able to run fully-managed Horovod jobs for