The AWS Deep Learning AMIs  for Ubuntu  and Amazon Linux  now come with an optimized build of TensorFlow 1.12  and support for MMS 1.0 , a Model Server for Apache MXNet  that provides a flexible and easy way to serve deep learning models exported from MXNet  or the Open Neural Network Exchange (ONNX) . For Amazon EC2 C5 instances, Deep Learning AMIs deploy compute-optimized TensorFlow built with Intel Advanced Vector Extensions (AVX instruction sets) to speed up the performance of vector and floating point operations. The AMIs also come pre-configured to leverage Intel Math Kernel Library for Deep Neural Networks (MKL-DNN). Training a ResNet-50 benchmark with the synthetic ImageNet dataset using our optimized build of TensorFlow 1.12 on a c5.18xlarge instance type was 13x faster than training using the stock TensorFlow 1.12 binaries.

Amazon Transcribe is an automatic speech recognition (ASR) service that makes it easy for you to add a speech-to-text capability to your applications. Amazon Transcribe now supports transcription of audio in the following new languages: British English, Australian English, and Canadian French. These languages expand upon the existing two languages already available in Amazon Transcribe: US English and US Spanish.

Amazon Comprehend for advanced text analytics now includes Custom Classification.

With the introduction of Amazon S3 Block Public Access, securing your S3 data has never been easier. With a few clicks in the S3 management console, you can apply S3 Block Public Access to every bucket in your account – both existing and any new buckets created in the future – and make sure that there is no public access to any object. By default, new S3 bucket settings do not allow public access, but customers can modify these settings to grant public access using policies or object-level permissions. The Amazon S3 Block Public Access settings override S3 permissions that allow public access, making it easy for the account administrator to set up a centralized control to prevent variation in security configuration regardless of how an object is added or a bucket is created. These settings are auditable, providing a further layer of control, using AWS Trusted Advisor bucket permission checks, AWS CloudTrail logs, Amazon Macie and Amazon CloudWatch.

AWS Secrets Manager makes it easier to follow the security best practice of using short-term secrets by rotating secrets safely on a schedule that you determine. For example, you can configure Secrets Manager to rotate a database credential daily, turning a typical, long-term secret in to a short-term secret that is rotated automatically. Now, Secrets Manager makes it easier for you to manage and rotate secrets by introducing three enhancements to the Secrets Manager console . First, you can tag secrets from the console. Second, you can give the rotation Lambda function a custom name. Finally, you can use existing Lambda functions to rotate new secrets.

This Quick Start builds an Internet of Things (IoT) Camera Connector environment and serverless architecture on the Amazon Web Services (AWS) Cloud in about 5 minutes.

Amazon Aurora Serverless (MySQL compatible edition) is now available in an additional 9 AWS Regions. With the additions of Asia Pacific (Mumbai), Asia Pacific (Seoul), Asia Pacific (Singapore), Asia Pacific (Sydney), US West (N. California), Canada (Central), EU (Frankfurt), EU (London), and EU (Paris), you can now choose the Serverless configuration of Amazon Aurora in 14 geographic Regions.

You can now retain Amazon RDS automated backups (system snapshots and transaction logs) when you delete a database instance. This allows you to restore a deleted database instance to a specified point in time within the backup retention period even after it has been deleted, protecting you against accidental deletion of data. 

The maximum payload size for asynchronous AWS Lambda functions is now 256KB. Previously, the limit was 128KB. Lambda supports multiple invocation modes today. Now, you can pass larger payloads when a function is invoked asynchronously, allowing Lambda to operate more seamlessly with services like SNS that already support larger payloads. Read our documentation  to learn more about the asynchronous invocation type. 

Amazon EC2 Fleet now supports a new request type, Instant, that lets you provision capacity synchronously across instance types, Availability Zones (AZs) and purchase models. The CreateFleet API will return the instances launched in the API response with the “Instant” request type. EC2 Fleet now offers 3 request types: instant, request, and maintain, that can be used to indicate whether EC2 Fleet should return instance information synchronously, launch instances asynchronously until the fleet reaches your desired capacity, or attempt to maintain your desired capacity.

When you set your fleet type to “Instant” you can rely on a synchronous API response, providing instance details including instance id, instance type and subnet rather than describing a fleet ID. When using Instant type fleets EC2 Fleet will not attempt to replenish instances if interrupted. Following the create-fleet synchronous response EC2 Fleet will take no further action enabling you to control if and when instances are launched.

Amazon EC2 Fleet is a feature that simplifies the provisioning of large amounts of EC2 capacity, including provisioning across different EC2 instance types, Availability Zones and across On-Demand, Reserved and Spot purchasing models. With a single API call, you can provision capacity that delivers the best mix of instance types and purchasing models to achieve your desired performance and cost. To learn more about EC2 Fleet, read the FAQs .