BACKGROUND IMAGE: iSTOCK/GETTY IMAGES
LAS VEGAS -- With its latest round of SageMaker updates, AWS continues its quest to make artificial intelligence more accessible to enterprise developers.
Amazon SageMaker is a managed service that enables developers to more easily fold machine learning (ML) capabilities into applications. AWS expanded the service here at re:Invent 2018 this week, with capabilities for data labeling, reinforcement learning (RL) and more.
"We've seen a lot of ML- and AI-related announcements from Amazon," said Sam Kroonenburg, CEO and co-founder of A Cloud Guru, a London-based cloud computing training and certification firm. "AWS is driving more high-level application services in the cloud with these offerings. They are making refinements in the AI space ... pushing the competition with Google and others on that front."
Among the capabilities coming to the Amazon platform is SageMaker Ground Truth, a service that aims to improve how AI labels data. It builds an active model that learns from human behavior to enable easier, more accurate and automatic data labeling. Labeling is a challenge, particularly at scale, where it could take an individual developer an inordinate amount of time to label a thousand images or documents.
With Ground Truth, Amazon looks to simplify that process, said Kathleen Walch, an analyst with Cognilytica in Ellicott City, Md.
In addition, SageMaker RL is a service designed to help developers and data scientists get involved in reinforcement learning, a type of machine learning where the system determines the actions it should take to reach an optimal outcome. Teams that want to add AI and machine learning capabilities to existing applications can do so with SageMaker, and then use SageMaker RL to simplify the incorporation of reinforcement algorithms into those apps, Walch said.
SageMaker Workflows targets collaboration and more
Other SageMaker updates deliver more automation, orchestration and collaboration to help developers build, manage and share machine language workflows, said Joel Minnick, senior manager of product marketing for AI and machine learning at AWS.
With SageMaker Workflows, developers can try new algorithms and iteratively test different machine learning models to find the one that suits their project's needs. Then, with SageMaker Search -- a feature within Workflows -- they can find other relevant models that are based on SageMaker model training runs in the AWS console.
Moreover, developers can connect source code management via GitHub, AWS CodeCommit or self-hosted Git repositories for enhanced collaboration and version control. Developers also can use AWS Step Functions to automate SageMaker workflows end to end. AWS Step Functions connect multiple AWS offerings into serverless workflows that underpin applications. SageMaker can then integrate with the open source Apache Airflow framework to create, schedule and monitor multistage workflows.
Additionally, Amazon SageMaker gains new algorithms and frameworks, including those to detect suspicious IP addresses. Developers can also bring their own custom container to run algorithms for model training in SageMaker; use built-in SageMaker algorithms; or manage MXNet, TensorFlow, PyTorch and Chainer algorithms, Minnick said.
These features to organize, track and evaluate machine learning model training experiments enable developers to monitor and automate the whole workflow process, Walch said. Moreover, companies need to continuously monitor for fraud. The ability to identify suspicious IP addresses enables them to be proactive and helps merchants make broader observations about fraud.
Machine-learning-as-a-service competitor Sift Science also offers AI algorithms for fraud detection and analysis, with IP addresses as one of the factors, Walch noted. "We see this as a good use of machine learning, as ML is very good at detecting patterns and anomalies," she said.
More AI additions
Amazon SageMaker is a software play, but AWS will also deliver high-performance machine learning inference chips, known as AWS Inferentia, next year. They are designed by AWS-owned Annapurna Labs in Cupertino, Calif.
The cloud provider also unveiled a new location to sell machine learning algorithms in the AWS Marketplace, a move intended to help users find the models they need and simply plug them in.
In addition, AWS has partnered with consultants to promulgate adoption of SageMaker to solve real-world problems. For instance, Dallas-based Pariveda Solutions has teamed up with AWS to use SageMaker to help food companies extract meaningful biological information from food matter, such as fruit and vegetables.
Meanwhile, Qubole, a data platform provider in Santa Clara, Calif., launched a service that enables developers and data scientists to use its platform and Apache Spark to cleanse and prepare data for machine learning models via SageMaker. Developers can use SageMaker to easily train and deploy the models. Qubole helps cut data prep time, while SageMaker accelerates the model training process.
Through a new AWS Training and Certification: Machine Learning offering, AWS has also made available all the machine learning courses it uses internally to train its engineers. There are more than 30 self-service courses that cover AWS machine learning services, including Amazon SageMaker, AWS DeepLens, Amazon Rekognition, Amazon Lex, Amazon Polly and Amazon Comprehend.