Register now for your free virtual pass to the Low-Code/No-Code Summit this November 9. Hear from executives from Service Now, Credit Karma, Stitch Fix, Appian, and more. Learn more.
Amazon SageMaker, which got its start five years ago, is among the most widely used machine learning (ML) services in existence.
Back in 2017 Sagemaker was a single service designed to help organizations use the cloud to train ML models. Much like how Amazon Web Services (AWS) has grown significantly over the last five years, so too has the number of ML services under the Sagemaker portfolio.
In 2018, Amazon SageMaker GroundTruth added data labeling capabilities. In 2019, AWS expanded SageMaker with a number of services including SageMaker Studio, which provides an integrated developer environment (IDE) for data scientists to build ML application workflows. The SageMaker Data Wrangler service was announced in 2020 for data preparation and in 2021 new capabilities included the Clarify explainability and ML feature store services.
AWS is continuing to add services to SageMaker, including a pair of announcements made yesterday, with new support for AWS Graviton cloud instances and multi-model endpoint support. During an AWS event on Oct. 26, Bratin Saha, VP and general manager of AI/ML at AWS, said there are over 100,000 customers from virtually every industry who make use of AWS’s cloud ML services.
Join today’s leading executives at the Low-Code/No-Code Summit virtually on November 9. Register for your free pass today.
“Machine learning isn’t the future that we need to plan for, it’s the present that we need to harness now,” Bratin said.
AWS scales SageMaker with multi-model endpoints (MME)
One of the things that has happened over the last five years of SageMaker adoption is an increase in scale for how models are trained and deployed.
To help organizations deal with the challenge of scaling, Bratin said that AWS has released the SageMaker multi-model endpoints (MME) capability.
“This allows a single GPU to host thousands of models,” Bratin said. “Many of the most common use cases for machine learning, such as personalization, require you to manage anywhere from a few hundred to hundreds of thousands of models.”
For example, Bratin said that in the case of a taxi service, an organization might have custom models based on each city’s traffic pattern. He noted that in a traditional machine learning system, a customer would have to deploy one model per instance, which means they would have to deploy hundreds or thousands of instances.
SageMaker MME changes that need, giving organizations the capability to host many models on a single instance, which lowers overall costs. Bratin said the MME service also handles all the work of orchestrating the ML model traffic and uses sophisticated caching algorithms to understand which model should be resident in memory at a particular time.
How one company continues to benefit from SageMaker
Among the many users of Amazon SageMaker services is Mueller Water Products.
Mueller Water Products is using Amazon SageMaker to help with its mission of limiting water loss. Using the ML service alongside its EchoShore-DX system for leak detection, the company has been able to achieve a 40% improvement in precision.
“AWS has really been able to consolidate various needs in the machine learning environments into one tool set which has been really efficient for our team to use,” Dave Johnston, director of smart infrastructure at Mueller Water Products, told VentureBeat.
Johnston said that many organizations, including utilities, have more data than they know what to do with. In his view, with the ML tools that AWS has developed in SageMaker, there is a lot of opportunity, not just for the water utility industry but for many different industries.
“There’s a lot of hidden value in the data that’s already been collected and there’s going to be lots of opportunities to unlock that value,” Johnston said. “I think [SageMaker is] a low-cost approach to unlock hidden value without having to deploy a bunch of new, expensive infrastructure and you can do it with data you’re already collected.”