Formerly known as Wikibon
Close this search box.

Deeper into cloud data: AWS launches a blitz of innovative AI offerings at re:Invent

Amazon Web Services Inc. made artificial intelligence far and away the dominant theme of its announcements on Wednesday, the second day of re:Invent 2018.

As presented in detail in AWS Chief Executive Andy Jassy’s keynotes, the latest announcements fell into the following categories:

  • Driving innovative AI into every cloud application;
  • Building and optimizing diverse data workloads in the cloud; and
  • Managing rich cloud-native applications across more complex deployments

Here are Wikibon’s dissection of the principal announcements from re:Invent 2018’s second day.

Driving innovative AI into every cloud application

AWS has introduced a staggering range of innovative new AI capabilities at this year’s re:Invent. These range from a new AI hardware-accelerator architecture, fully managed cloud services for diverse enterprise AI use cases, and even a prototype miniature autonomous vehicle powered by a cutting-edge AI modeling and training methodology.

On day two at re:Invent, the principal AI-related announcements were as follows:

  • Introducing a new hardware architecture for fast AI inferencing in the cloud: The vendor announced development of AWS Inferentia, a new high-performance AI-accelerator chip that will be available some time in 2019. AWS is engineering the chip for high throughput, low latency AI apps as an alternative to a GPUs. AWS claims that it provides “order of magnitude” cost reduction when running inferencing workloads in the AWS cloud. When deployed in parallel into the AWS cloud, Inferentia hardware will “scale to thousands of TOPS” [tera operations per second]. It will be able to execute AI models built in TensorFlow, Mxnet and PyTorch. It will support hundreds of teraflops per chip and thousands of teraflops per Amazon EC2 instance for multiple frameworks and multiple data types. It is optimized for multiple data types, including INT-8 and mixed precision FP-16 and bfloat16. And it will work with Amazon EC2, SageMaker and Elastic Inference.
  • Optimizing the dominant AI development framework for its cloud: The vendor announcedgeneral availability of the AWS-Optimized TensorFlow framework. AWS has boosted TensorFlow’s scalability across GPUs, thereby accelerating training and inferencing of models built in TensorFlow when those workloads run inside AWS’ cloud. AWS claims that it has optimized TensorFlow to achieve linear scalability when training several types of deep learning algorithms and other neural networks. When used with the newly announced P3dn instances, AWS-Optimized TensorFlow has achieved 90 percent efficiency across 256 GPUs, as compared with the previous 65 percent efficiency.
  • Enabling fast access to the best AI algorithms and pretrained models: The vendor announced general availability of AWS Marketplace for Machine Learning. Supplementing the popular models and algorithms that are bundled with Amazon SageMaker, the marketplace gives developers access over 150 additional algorithms and pretrained models, with new ones added daily. All of these can be deployed directly into SageMaker for immediate use by developers. Developers use the marketplace’s self-service interface to list and sell their own algorithms and models through that site.
  • Automating labor-intensive labeling of AI training data: The vendor announced general availability of Amazon SageMaker Ground Truth. This new solution enables developers to automate low-cost, high-throughput, highly accurate labeling of training data using human annotators through Mechanical Turk, third party vendors, or their own employees. The solution uses AI to learn from these annotations in real time and can automatically apply labels to much of the remaining dataset, thereby reducing the need for human review of the labeled data prior to its use in training AI models in SageMaker.
  • Scaling, speeding and reducing the cost of fast AI inferencing in the cloud: The vendor announced general availability of Amazon Elastic Inference. This new fully managed service enables AI developers to run inferencing workloads on a general-purpose Amazon EC2 instance and provision just the right amount of GPU performance. Starting at just 1 TFLOP, developers can elastically increase or decrease the amount of inference performance, and only pay for what they use. Elastic Inference enables significant cost savings on cloud-based inferencing workloads, when compared to inferencing on a dedicated Amazon EC2 P2 or P3 instance with relatively low utilization. Elastic Inference supports all popular AI frameworks (including TensorFlow, PyTorch and MXNet) and is integrated with Amazon SageMaker and the Amazon EC2 Deep Learning Amazon Machine Image. Developers can start using Amazon Elastic Inference immediately without making any changes to their existing models.
  • Accelerating AI inferencing automatically to disparate edge devices: The vendor announced general availability of Amazon SageMaker Neo. This new AI model compiler lets customers train models once and run them anywhere with, according to AWS claims, up to 2x performance improvements. It compiles AI models for specific target hardware platforms and optimizes their performance automatically without compromising model accuracy. It thereby eliminates the need for AI developers. To manually tune their trained models for each target hardware platform. It currently supports AI hardware platforms from NVIDIA, Intel, Xilinx, Cadence, and Arm, as well as popular frameworks such as TensorFlow, Apache MXNet and PyTorch. AWS also indicated that it plans to make Neo available as an open source project.
  • Bringing reinforcement learning into mainstream AI initiatives: The vendor announcedgeneral availability of Amazon SageMaker RL, which is the cloud’s first managed reinforcement learning service for machine learning development and training pipelines. The new fully managed service enables any SageMaker user to build, train, and deploy machine learning models through any several built-in RL frameworks, including Intel Coach and Ray RL and to leverage any of several simulation environments, including SimuLink and MatLab. It integrates with the newly announced AWS RoboMaker managed service, which provides a simulation platform for RL on intelligent robotics projects. It also works with the OpenGym RL environment, supports Amazon’s Sumerian mixed-reality solution, and interoperates with the open source Robotics Operating System.
  • Delivering AI-personalized recommendations into cloud applications: The AWS announced limited preview of Amazon Personalize, a fully managed service that uses AI for generate real-time recommendations. Incorporating recommender technology that is used operationally in’s online retailing business, the new service supports building, training, and deployment of custom, private personalization and recommendation models for virtually any use case. Amazon Personalize can make context-aware, personalized recommendations and segment customers for 1:1 marketing through Web, email and other channels and user experience models. It leverages automated machine to continuously learn and tune its recommendations to maximize results. It keeps data private and encrypted and incorporates algorithms and models that are built and trained in Amazon SageMaker.
  • Automating delivery of AI-generated time-series forecasting into cloud applications: The company announced limited preview of Amazon Forecast. Incorporating forecasting technology that is used operationally in’s online retailing business, the new fully managed service uses AI creates accurate time-series forecasts. It uses historical time-series data to automatically train, tune, and deploy custom, private machine learning forecasting models. It uses automated machine learning to work with any historical time series and even analyze multiple time series at once. It provides forecast visualization and can import results into business apps. It can incorporate existing machine learning algorithms built and trained in Amazon SageMaker.
  • Performing high-volume AI-driven OCR on any document in the cloud: The vendor announced limited preview of Amazon Textract, a new fully managed service that uses AI to instantly read virtually any type of document and accurately extract text and data without need for manual reviews or custom coding. It incorporates optical character recognition and enables developers, without any AI or machine learning skills, to quickly automate document workflows in order to process millions of document pages in a few hours.
  • Leveraging AI to extracting medical data rapidly from diverse file and data stores and formats: The vendor announced general availability of Amazon Comprehend Medical, a new fully managed service that can extract medical data quickly from virtually any document. The service applies natural language processing to medical text, using machine learning to extract disease conditions, medications and treatment outcomes from patient notes, clinical trial reports, and other electronic health records. It requires no machine learning expertise, no complicated rules to write, no models to train and is continuously improving.
  • Encouraging development of RL-based robotics for autonomous edge devices: The vendor announced that AWS DeepRacer is in limited preview and is now available for pre-order. DeepRacer is a fully autonomous toy race car that, though one-eighteenth the scale of a real one, comes equipped with all-wheel drive, monster truck tires, a high-definition video camera, and on-board compute). What drives is in AI model that was built and trained in RL algorithms, workflows, and a simulator included with SageMaker RL. Developers only need a few lines of code to start learning about RL through DeepRacer. Developers can benchmark their DeepRacer cars and the embedded RL models against each other in what AWS refers to as “the world’s first global autonomous racing league.”

Building and optimizing diverse data workloads in the cloud

AWS continued to roll out new and enhanced cloud data platforms, following on the many announcements it made in that regard on the day prior. Many of the latest announcements involved enhancements in the price, performance, accessibility, availability, and scalability of AWS’ existing cloud data platforms, though it did roll out new data platforms for immutable hyperledgers.

On day two at re:Invent, the principal data platform announcements were as follows:

  • Managing diverse file systems in the cloud: The vendor announced general availability of the Amazon FSx family, which includes two new fully managed third-party file system services that provide native support for Windows and compute-intensive workloads (using Lustre). It also introduced a new Infrequent Access storage class for Amazon Elastic File System (EFS), its file system service for Linux-based workloads. The new Amazon EFS Infrequent Access, which will be available in early 2019, allows users to reduce storage costs by up to 85 percent compared to the Amazon EFS Standard storage class. With EFS IA, Amazon EFS customers only need to enable Lifecycle Management to automate the movement to this new storage class of any file that has not been accessed in more than 30 days.
  • Rapidly building secure data lakes in the cloud: The vendor announced limited preview of AWS Lake Formation, a fully managed service to simplify and accelerate the setup of secure data lakes. AWS Lake Formation allows users to define the data sources they wish to ingest and then select from a prescribed list of data access and security policies to apply. This remove the need to define and enforce policies across their various analytics applications that use the data lake. The service then collects the data and moves it into a new Amazon S3 data lake, extracting technical metadata in the process to catalog and organize the data for easier discovery. It automatically optimizes the partitioning of data to improve performance and reduce costs, transforms data into formats like Apache Parquet and ORC for faster analytics, and also uses machine learning to deduplicate matching records to increase data quality. It supports central definition and management of security, governance, and auditing policies for the data lake. It also provides a centralized, customizable catalog which describes available data sets and their appropriate business use.
  • Running a robust, high-performance global relational database in the cloud: The vendor announced general availability of Amazon Aurora Global Database. This enables users to update Aurora in a single AWS Region and automatically replicate the update across multiple AWS Regions globally in less than a second. This enhancement to AWS’ fully managed cloud relational database service enables users to maintain read-only copies of their database for fast data access in local regions by globally distributed applications, or to use a remote region as a backup option in case they need to recover their database quickly for cross-region disaster recovery scenarios.
  • Managing a global key-value database cost effectively and with transactional guarantees: The vendor announced general availability of DynamoDB On-Demand. This enhancement to AWS’ fully managed, key-value database service offers reliable performance at any scale. For applications with unpredictable, infrequent usage, or spikey usage where capacity planning is difficult, Amazon DynamoDB On-Demand removes the need for capacity planning. It automatically manages read/write capacity, and users only pay-per-request for the cloud resources that they actually use. AWS also announcedDynamoDB Transactions, a new fully managed services that enables developers to easily build transactional guarantees of full atomicity, consistency, isolation, and durability into multi-item updates in their DynamoDB applications.
  • Automating bulk cloud storage management: The company announced Amazon S3 Batch Operations. This new service, which will be available in early 2019, automates management of thousands, millions, or billions of data objects in bulk storage. It enables developers and IT administrators to change object properties and metadata and execute storage management tasks for large numbers of Amazon S3 objects with a single API request or a few clicks in the Amazon S3 Management Console.
  • Archiving data securely and inexpensively in the cloud: The company announced Amazon S3 Glacier Deep Archive, which will be available in early 2019. Designed as an alternative to tape infrastructure, this is a new secure storage class for users to archive large data sets cost-effectively while ensuring that their data is durably preserved for future use and analysis.
  • Performing low-latency streaming time-series and event data analytics in the cloud: The company announced the preview of Amazon Timestream, a new fully managed time series database service. AWS claims that Amazon Timestream processes and analyzes trillions of events per day at one-tenth the cost of relational databases, with up to one thousand times faster query performance than a general-purpose relational database. It includes such AI-driven analytics functions as smoothing, approximation, and interpolation to help customers identify trends and patterns in real-time data. Its serverless, architecture automatically scales up or down to adjust capacity and performance, so that users only pay for the cloud resources that they consume.
  • Running an append-only immutable hyperledger distributed database: The vendor announced the preview of Amazon Quantum Ledger Database. This is a fully managed hyperledger clouid database service. It is serverless, immutable, scalable, and cryptographically verifiable. AWS claims that is has a 2-3x transaction-processing capacity than blockchain-based hyperledgers, owing to the fact that it doesn’t require distributed consensus to make updates. In addition, the vendor announced limited preview of AWS Managed Blockchain. This is a fully managed service that makes it easy to quickly create and manage scalable blockchain networks to transact and securely share data. AWS supports both the Hyperledger Fabric and Ethereum blockchain platforms.

Managing rich cloud-native applications across more complex deployments

AWS continued to deepen its support for management of cloud computing application deployments in hybrid, multi-, and edge cloud scenarios. On day two at re:Invent, the principal cloud platform management announcements were as follows:

  • Managing cloud-native services transparently across hybrid clouds: The vendor announced AWS Outposts. In private preview with general availability expected in the second half of 2019, these are fully managed and configurable compute and storage racks. They incorporate AWS-designed hardware and enable customers to run compute and storage on-premises while seamlessly connecting to the rest of AWS’s public cloud services. On the user’s premises, AWS Outposts run services as Amazon EC2 and EBS. Customers who want to use the same VMware control plane and APIs they’ve been using to run their on-premises infrastructure can run VMware Cloud on AWS locally on AWS Outposts. They can also manage it as a service from the same console as VMware Cloud on AWS. Customers who prefer the same APIs and control plane they use in AWS’s cloud can use the AWS-native variant on premises with AWS Outposts. AWS Outposts can also run VMware services such as NSX, AppDefense, and vRealize Automation across VMware and Amazon EC2 environments. Either way, AWS delivers the racks to customers, installs them, and handles all maintenance and replacement of the racks. They are an extension of a user’s Amazon virtual private cloud in the closest AWS Region to each customer.
  • Managing multiple cloud accounts from a single location: The vendor announced limited preview of AWS Control Tower. This is a fully managed services that makes it easy to configure and govern a secure, compliant multi-account AWS environment It provides cloud teams with a single, automated “landing zone” where their teams can provision accounts and workloads. It provides curated guardrails for policy enforcement, employing best-practices blueprints, such as configuring a multi-account structure using AWS Organizations, managing user identities and federated access with AWS Single Sign-on or Microsoft Active Directory, configuring an account factory through AWS Service Catalog, and centralizing a log archive using AWS CloudTrail and AWS Config. It provides pre-packaged governance rules for security, operations, and compliance. It supports easy monitoring and management of all this through a dashboard that provides continuous visibility into a customer’s AWS environment.
  • Centralizing cloud security: The vendor announced the preview of AWS Security Hub, a fully managed service that provides centralized management of a user’s cloud security and compliance. It allows users to quickly see their entire AWS security and compliance state in a central location. It collects and aggregates findings from the security services it discovers in a customer’s environment, such as intrusion detection findings from Amazon GuardDuty, vulnerability scan results from Amazon Inspector, sensitive data identifications from Amazon Macie, and findings generated by a range of security tools from AWS Partner Network partners. It correlates these findings into integrated dashboards that visualize and summarize a customer’s current security and compliance status, and also highlight trends. Users can run automated, continuous configuration and compliance checks based on industry standards and best practices. It integrates with Amazon CloudWatch and AWS Lambda, thereby enabling users to execute automated remediation actions based on specific types of findings.

To catch what AWS executives, partners and customers are saying now, get drill-downs on their forthcoming announcements and receive compelling glimpses into their roadmaps, be sure to tune into theCUBE live this week.

Book A Briefing

Fill out the form , and our team will be in touch shortly.
Skip to content