At the AWS re:invent conference 2017, held in Las Vegas, USA, several new AWS machine learning (ML) and Internet of Things (IoT) products were released. Highlights include Amazon SageMaker - a fully-managed ML service that enables developers to "quickly build, train, and host machine learning models at scale"; AWS DeepLens - an ML-enabled edge-device with a camera; AWS IoT One-Click; AWS IoT Device Defender; and IoT Device Manager - a service to securely onboard, organise, monitor, and remotely manage IoT devices at scale.
Andy Jassy, CEO of Amazon Web Services (AWS), began the second half of the day one keynote by stating that machine learning is an extremely valuable technique that is relevant to many areas of an enterprise organisation. However, modern ML frameworks are often challenging for developers without specialist knowledge to work with. Jassy proposed that ML can be consumed when developing software applications within a three layer model: frameworks and interfaces - Apache MXNet, TensorFlow, PyTorch etc; platform services - this is where the new AWS product launches would focus; and application services - Amazon Rekognition, Amazon Polly, Amazon Lex etc.
Jassy continued by stating that ML platform services would allow more developers to leverage the benefits of ML, particularly if there was integration with these services with the wider cloud platform and ecosystem. The first product announcement was Amazon SageMaker - a fully-managed end-to-end machine learning service that enables data scientists, developers, and machine learning experts to "quickly build, train, and host machine learning models at scale". There are three main components to Amazon SageMaker: Authoring - zero-setup hosted Jupyter notebook IDEs for data exploration, cleaning, and preprocessing; model training - a distributed model building, training, and validation service; and model hosting - a model hosting service with HTTPS endpoints for invoking models to get real-time inferences.
Despite decades of investment and improvements, the process of developing, training, and maintaining machine learning models has still been cumbersome and ad-hoc [...] Businesses and developers want an end-to-end, development to production pipeline for machine learning.
The next ML product launch was AWS DeepLens - a video camera that runs deep learning ML models directly on the device. This first ML-enabled edge device from AWS includes a 4 megapixel camera and a 2D microphone array. The device is powered by an Intel Atom Processor that provides over 100 GFLOPS of compute power and 8 gigabytes of memory for pre-trained models and code. AWS DeepLens runs Ubuntu 16.04 and is preloaded with the Greengrass Core (Lambda runtime, message manager etc), and a device-optimised version of MXNet. There is the flexibility to use other frameworks such as TensorFlow and Caffe2.
Amazon Rekognition Video is a new video analysis service that brings scalable computer vision analysis to S3 stored video, as well as live video streams. Amazon Rekognition Video allows the "accurate detection, tracking, recognition, and extraction of thousands of objects, faces, and content from a video". This service builds on last year's announcement of Amazon Rekognition Image, which enables engineers to build and integrate object and scene detection, real-time facial recognition, image moderation, as well as text recognition into applications.
A private preview of Amazon Transcribe was also launched, which provides an automatic speech recognition (ASR) service that allows engineers to add speech-to-text capabilities to their applications. Audio files stored on S3 in many common formats (WAV, MP3, Flac, etc.) can be analysed by starting a job with the Transcribe API. The output of the service is a detailed transcription with timestamps for each word, as well as inferred punctuation. The preview release provides an asynchronous transcription API to transcribe speech in English or Spanish.
Keeping within the topic of text analysis, Jassy continued with the GA release of Amazon Comprehend - a service that analyses text and extracts the language "from Afrikans to Yoruba, with 98 more in between", and for English and Spanish can identify different types of entities (people, places, brands etc), key phrases, sentiment (positive, negative, mixed, or neutral), and extract key phrases. The first four functions -- language detection, entity categorisation, sentiment analysis, and key phrase extraction -- are designed for interactive use, with responses available in hundreds of milliseconds. Comprehend is a continuously-trained trained Natural Language Processing (NLP) service, and the goal is to make the service increasingly accurate and more broadly applicable over time.
In the final section of the keynote Jassy changed focus slightly, and mused on the emerging value of "Internet of Things (IoT)" to the enterprise. IoT devices are now pervasive, and the density of their deployment is predicted to increase significantly over the coming years: from light bulbs to industrial machinery; and from "smart cars" to "smart cities". Jassy stated that AWS is focused on helping enterprises take advantage of this new trend.
The first product release within the IoT domain was AWS IoT 1-Click, a service that makes it easy for simple devices -- such as single-purpose, cloud-connected devices like badges or buttons -- to trigger AWS Lambda functions that execute a specific action. Examples of possible actions include calling technical support, reordering goods and services, or locking and unlocking doors and windows. The AWS IoT Enterprise Button, based on the Amazon Dash Button hardware, is a simple device that can be purchased and used with the AWS IoT 1-Click service. A mobile app can be downloaded from Google Play or the Apple App Store that allows a list of devices that are already pre-configured to securely connect to an AWS account. After selecting a device, a customer chooses the Lambda function they want to run when the device sends the trigger.
Several additional new products were launched within the IoT space, all in private preview. AWS IoT Device Management can be used to securely onboard, organise, monitor, and remotely manage (including updating device software over-the-air) IoT devices at scale throughout their lifecycle. AWS IoT Device Defender is a fully-managed service that allows customers to secure a fleet of IoT devices on an ongoing basis. AWS IoT Device Defender audits a fleet to ensure it adheres to security best practices, detects abnormal device behavior, and creates alerts for security issues.
AWS IoT Analytics is a service that provides "advanced data analysis" of data collected from IoT devices. The AWS IoT Analytics service will allow customers to process messages, gather and store large amounts of device data, as well as query data. AWS IoT Analytics also integrates with Amazon QuickSight for the visualisation of data, and allows experimentation with machine learning through an integration with Jupyter Notebooks.
Related to the previous announcement of DeepLens, a new AWS Greengrass ML Inference service makes it easy to perform ML inference locally on AWS Greengrass devices using models that are built and trained in the cloud. The AWS blog states that this will enable engineers to push ML out towards the edge:
Until now, building and training ML models and running ML inference was done almost exclusively in the cloud [...] With AWS Greengrass ML Inference your AWS Greengrass devices can make smart decisions quickly as data is being generated, even when they are disconnected.
AWS Greengress ML Inference simplifies each step of working with ML, including accessing ML models, deploying models to devices, building and deploying ML frameworks, creating inference apps, and utilising on-device accelerators such as GPUs and FPGAs. The AWS blog states that it is possible to "access a deep learning model built and trained in Amazon SageMaker directly from the AWS Greengrass console and then download it to your device as part of an AWS Greengrass group". The service also includes a prebuilt Apache MXNet framework to install on AWS Greengrass devices.
The final announcement in this space was Amazon FreeRTOS, an IoT microcontroller operating system that simplifies development, security, deployment, and maintenance of microcontroller-based edge devices. Amazon FreeRTOS extends the FreeRTOS kernel, a popular real-time operating system, with libraries that enable local and cloud connectivity, security, and (coming soon) over-the-air updates.
Additional product releases and announcements made at AWS re:invent 2017 can be found on the AWS news blog.