AWS adds governance, geospatial tools to Amazon SageMaker
The cloud computing giant added data governance and geospatial tools to its machine learning platform and provided 40 new data connectors to Data Wrangler.
AWS updated its Amazon SageMaker machine learning platform on Wednesday with new data governance features and geospatial data tools and added some 40 data connectors to the platform's Data Wrangler module.
The cloud giant has steadily modernized SageMaker since it introduced the system five years ago and has seen it become one of the most widely used -- though not the cheapest -- ML platforms for enterprises.
"Enterprise AI initiatives are accelerating in maturity and are actively devising operational AI systems that integrate data and AI pipelines," said Chirag Dekate, an analyst at Gartner. "The SageMaker announcements today address many of these concerns, including collaboration, role and model management, enabling diverse teams to integrate their expertise to deliver high-efficiency operational AI systems."
Among the eight new capabilities added to SageMaker, the vendor significantly strengthened data governance within the low-code ML platform, said Swami Sivasubramanian, vice president of databases, analytics and machine learning at AWS, during a keynote on Wednesday at the AWS re:Invent 2022 conference.
Amazon SageMaker Role Manager defines custom permissions and comes with predefined policy templates for various personas and ML activities, according to AWS.
Capabilities like model dashboards simplify management of multiple AI pipelines and improve observability across the operational AI system.
Chirag DekateAnalyst, Gartner
SageMaker Model Cards aim to streamline model documentation during the ML lifecycle by creating a single source of truth for information about models and autopopulating details such as training jobs and data sets, model artifacts, and inference environment, according to the vendor.
And a new SageMaker Model Dashboard monitors the performance of all models in one place.
"These are really powerful governance capabilities that will help you build ML governance responsibility to address power permission sharing," Sivasubramanian said during the streamed keynote.
"Capabilities like model dashboards simplify management of multiple AI pipelines and improve observability across the operational AI system," Dekate said.
Geospatial tools
Meanwhile, the new geospatial features extend the range of SageMaker's ML model building capabilities into the burgeoning geospatial engineering sector, for applications such as disaster response, urban planning, logistics and transportation.
AWS introduced the geospatial tools -- designed to simplify building, training and deploying complex geospatial models -- in preview in the U.S. West region.
The geospatial collection provides pre-trained deep neural network models and geospatial operators that enable users to easily connect to and prepare large geospatial data sets, according to AWS. Predictions generated by models can be visualized on maps.
Introducing geospatial tools to Amazon SageMaker dovetails with AWS' pronounced focus on supply chain technology, said Bradley Shimmin, an Omdia analyst.
And some of the other updates reflect AWS incorporating what some of its independent competitors in the AI and ML arena have already specialized in, such as Databricks with data prep and integration, and DataRobot with model testing, he said.
"For a long time, they've been working to make SageMaker a platform that can accommodate a wide array of data types, and what you saw with the geospatial is absolutely the right direction," Shimmin said. "It's something that they're not the only ones to do it, and if you look down the list of features and functions they've added to SageMaker today, you'll see a lot of the same sort of ideas that have been popping up within smaller competitors for some time now."
Data connectors to the outside world
Sivasubramanian noted that the several dozen new data connectors for Amazon SageMaker complement 22 other new connectors AWS introduced Wednesday for platforms including LinkedIn and Google Ads, and other applications such as Snowflake, Salesforce and SAP.
"Today, we are bringing you more than 40 new connectors through SageMaker Data Wrangler, allowing you to implement and input even more of your data for ML model building and training," Sivasubramanian said.
AWS competes with many of those cloud platforms, and the vendor has been faulted for being slow to provide support for third-party applications, but Shimmin said it's unlikely that AWS is opening up in the face of criticism.
Rather, "it's a recognition that within data science in particular, companies are very willing to utilize a multivendor solution, because they want to use the best tool to solve the problem they have at hand," Shimmin said.
