Review: AWS AI and Machine Learning stacks up, and up

Amazon Fraud Detector

Amazon Fraud Detector is a managed service that allows you to identify potentially fraudulent online activities, such as online payment fraud and the creation of fake accounts. Fraud Detector uses automatic machine learning and 20 years of fraud detection expertise from AWS and Amazon.com to automatically identify potentially fraudulent activity.

There are five steps to using Amazon Fraud Detector:

1. Define the event you want to evaluate for fraud.
2. Upload your historical event dataset to Amazon S3 and select a fraud detection model type.
3. Amazon Fraud Detector uses your historical data as input to build a custom model. The service automatically inspects and enriches data, performs feature engineering, selects algorithms, trains and tunes your model, and hosts the model.
4. Create rules to either accept, review, or collect more information based on model predictions.
5. Call the Amazon Fraud Detector API from your online application to receive real-time fraud predictions and take action based on your configured detection rules.

IDG

Amazon Lookout for Vision (Preview)

Amazon Lookout for Vision is a machine learning service that spots defects and anomalies in visual representations using computer vision. For example, Amazon Lookout for Vision can be used to identify missing components in products, damage to vehicles or structures, irregularities in production lines, minuscule defects in silicon wafers, and other similar problems.

Lookout for Vision doesn’t require specialized machine vision cameras. It works with five steps:

1. Collect images that show normal and defective products from your production line and load them into the Amazon Lookout for Vision console.
2. Label images as normal or anomalous and Lookout for Vision will automatically build a model for you in minutes. Tune your model to improve defect detection by adding images to the dataset.
3. Use the Amazon Lookout for Vision dashboard to monitor defects and improve processes.
4. Automate visual inspection processes in real time or in batch and receive notifications when defects are detected.
5. Make continuous improvements by providing feedback on the identified product defects.

At this point, Lookout for Vision sends visual inspection images to the AWS cloud for classification. You can tighten the loop if you set up AWS Outposts servers on-premises so that you can run AWS services locally. But also consider AWS Panorama for computer vision at the edge.

IDG

AWS Panorama (Preview)

While Amazon Lookout for Vision is a cloud service tied to local cameras, AWS Panorama was designed to connect to cameras and provide computer vision at the edge. Panorama is a machine learning appliance and SDK that allows organizations to bring computer vision to on-premises cameras to make predictions locally with high accuracy and low latency.

The AWS Panorama Appliance is a hardware device that allows you to add computer vision to your IP cameras that weren’t built to accommodate computer vision. AWS Panorama Appliance turns your existing cameras into smart cameras that can run computer vision models on multiple concurrent video streams.

The AWS Panorama Device SDK is a software kit that enables third-party manufacturers to build new cameras that run more meaningful computer vision models at the edge for tasks like object detection or activity recognition. AWS Panorama-compatible cameras work out of the box with AWS machine learning services.

Typically you would train your computer vision models on SageMaker, and then run inference at the edge, either on a Panorama Appliance or Panorama-compatible cameras. AWS Panorama can import trained computer vision models from S3, and it uses SageMaker Neo to optimize models to run on the AWS Panorama Appliance.

Amazon Rekognition

Amazon Rekognition is a managed image and video analysis service. It can identify objects, people, text, scenes, and activities in images and videos, as well as detect any inappropriate content (content moderation). Rekognition also provides facial analysis and facial search capabilities that you can use to detect, analyze, and compare faces.

If Rekognition doesn’t do what you need, you can use Rekognition Custom Labels to identify the objects and scenes in images that are specific to your business needs. You need to supply images of objects or scenes you want to identify, and the service takes care of the transfer learning to build a customized model.

In June 2020, Amazon implemented a one-year moratorium on police use of facial recognition. The justification for the one-year period is that it would give legislators time to catch up to technology.

IDG

Amazon Polly

Amazon Polly is a managed text-to-speech service that includes a selection of standard and neural and male and female voices for over 20 languages and variants. By comparison, Microsoft Azure text-to-speech supports over 50 languages and variants, and Google text-to-speech supports over 40. The pricing of all three services is very close.

The neural voices, produced using deep learning, sound considerably more natural than the standard voices, but cost four times as much to use. A few neural voices are available for newscaster and conversational styles. Also, a few voices have been trained for bilingual use.

Amazon Polly supports Speech Synthesis Markup Language (SSML) with some custom Amazon tags. It supports MP3, Vorbis, and raw PCM audio stream formats at a range of sampling rates from 8 kHz, which sounds pretty bad, to 24 kHz, which is lifelike. You can request speech marks that signify the beginnings of words and sentences to help with highlight or animation synchronization; speech mark requests cost as much as speech requests.

Amazon Transcribe

Amazon Transcribe is a managed pay-as-you-go automatic speech recognition service based on deep learning. (I’ll discuss the specialized version, Transcribe Medical, with the other industry-specific services below.) Transcribe can be used to transcribe customer service calls, automate subtitling, and generate metadata for media assets to create a fully searchable archive.

Amazon Transcribe automatically adds speaker diarization, punctuation, number normalization, and formatting. In my experience, it doesn’t always get these right, especially if the speaker pauses for a breath or stumbles over a word. While there’s certainly a use for Transcribe, it doesn’t quite replace manual transcription services.

There are ways to improve your automatic transcription, however. You can add new words to the base vocabulary to generate more accurate transcriptions for domain-specific words and phrases such as product names, technical terminology, and the names of individuals. You can also specify a list of words to remove from transcripts, such as “uh,” “um,” and profanity.

Amazon Lex

Amazon Lex is a managed service for building conversational interfaces into any application using voice and text. It relies on Amazon Transcribe for voice recognition, and Amazon Comprehend to recognize the intent of the text. You can use Lex to build bots to increase contact center productivity and automate simple tasks. Lex uses the same technologies that power Alexa.

Amazon Lex can call AWS Lambda functions to do things like look up prices and make reservations. It can also support multi-turn conversations, so that it can fill in parameters that turn an initial intent such as “I want to book a room” to a specific request such as “Book a non-smoking room with a king-size bed at the Boston Marriott Long Wharf for two adults on Thursday night at the AARP discount rate.”

Amazon DevOps Guru (Preview)

There are few organizations that have as much experience with application operations as AWS and Amazon.com. Amazon DevOps Guru is a managed service that detects behaviors that deviate from normal operating patterns so you can identify operational issues before they impact your customers. DevOps Guru uses machine learning models trained on internal AWS operational data to provide accurate operational insights for critical issues that impact applications.

IDG

Amazon CodeGuru Reviewer and Profiler (Preview)

Amazon CodeGuru has two components: Reviewer to improve your code quality, and Profiler to identify an application’s most expensive lines of code.

Amazon CodeGuru Reviewer doesn’t necessarily replace human peer review, but augments it. You add Reviewer to your repository as one of the designated code reviewers. It works for Java and Python, initially analyzes whole repositories, and then analyzes pull requests as they come in, using machine learning, AWS and security best practices, and hard-learned lessons across millions of code reviews for thousands of open-source and Amazon repositories.

Amazon CodeGuru Profiler optimizes performance for applications running in production and identifies the most expensive lines of code. It is always searching for application performance optimizations, recommending ways to fix them to reduce CPU utilization, cut compute costs, and improve application performance.

AWS Industry Specific Solutions

There’s actually significant overlap between industry-specific solutions and standard AI services. AWS Panorama and Amazon Lookout for Vision, for example, could easily be classed in both categories — and AWS does that in its marketing materials. Nevertheless, I have included Amazon Monitron, Amazon Lookout for Equipment, Amazon Healthlake, Amazon Comprehend Medical, and Amazon Transcribe Medical in this category.

Amazon Monitron

Amazon Monitron is an end-to-end system that uses machine learning to detect abnormal behavior in industrial machinery, enabling you to implement predictive maintenance and reduce unplanned downtime. Monitron includes sensors to capture vibration and temperature data from equipment, a gateway device to securely transfer data to AWS, the service that analyzes the data for abnormal machine patterns using machine learning, and a companion mobile app to set up the devices and receive reports on operating behavior and alerts to potential failures in your machinery. This is the same technology used to monitor equipment in Amazon Fulfillment Centers.

Amazon Lookout for Equipment (Preview)

Amazon Lookout for Equipment uses the data from your existing sensors to detect abnormal equipment behavior, so you can take action before machine failures occur and avoid unplanned downtime. It analyzes the data from your sensors, such as pressure, flow rate, RPMs, temperature, and power, to automatically train a specific machine learning model based on just your data, for your equipment, with no machine learning expertise required on your end.

Amazon Healthlake (Preview)

Amazon HealthLake is a HIPAA-eligible service that enables healthcare providers, health insurance companies, and pharmaceutical companies to store, transform, query, and analyze health data at petabyte scale. It removes the heavy lifting of organizing, indexing, and structuring patient information to provide a complete view of the health of individual patients and entire patient populations in a secure, compliant, and auditable manner.

Using the Amazon HealthLake APIs, healthcare organizations can easily copy health data in the Fast Healthcare Interoperability Resources (FHIR) industry standard format from on-premises systems to a secure data lake in the cloud. HealthLake transforms unstructured data using specialized machine learning models, like natural language processing, to automatically extract meaningful medical information from the data and provides powerful query and search capabilities.

Amazon Comprehend Medical

Amazon Comprehend Medical is a HIPAA-eligible natural language processing (NLP) service that uses machine learning to extract health data from medical text, with no machine learning experience required. With an API call to Amazon Comprehend Medical you can extract information such as medical conditions, medications, dosages, tests, treatments and procedures, and protected health information, while retaining the context of the information.

Amazon Comprehend Medical can identify the relationships among the extracted information to help you build applications for use cases like population health analytics, clinical trial management, pharmacovigilance, and summarization. You can also use Comprehend Medical to link the extracted information to medical ontologies such as ICD-10-CM or RxNorm to help you build applications for revenue cycle management (medical coding), claim validation and processing, and electronic health record creation.