AWS Managed AI Services Flashcards
AWS AI Managed Services
Pre-trained ML services for your various use cases
Specialized CPU and GPUs for specific use-cases for cost saving; deployed across multiple AZs and AWS regions
Token-based pricing, pay for what you use; provisioned throughput for predictable workloads, cost savings and predictable performance
Comprehend
AWS ML service for Natural Language Processing; fully managed and serverless service
Uses machine learning to find insights and relationships in text
Get language of the text, extract key phrases/people/etc., understand text positivity/negativity, analyze text using tokenization
Analyze customer interactions to understand sentiment, create and group articles by topics this service will uncover
Custom Classification
Comprehend feature for organizing documents into categories that you define
For example, categorize customer emails so that you can provide guidance based on the type of the customer request
Supports different document types, such as text, PDF, Word, images, etc.
Real-Time Analysis for single documents, synchronous; Async Analysis for multiple documents in a batch, asynchronous
Custom Entity Recognition
Comprehend feature for analyzing text for specific terms and noun-based phrases
Extract terms like policy numbers, or phrases that imply a customer escalation, anything specific to your business
Train the model with custom data such as a list of the entities and documents that contain them
Real-Time or Async Analysis
Comprehend Medical
AWS service for detecting and returning useful information in unstructured clinical text
Uses NLP to detect PII; physician’s notes, discharge summaries, test results, case notes
Store documents in S3 and analyze real-time data with Kinesis Data Firehose
Use Transcribe Medical to transcribe patient narratives into text to be analyzed by this service
Translate
AWS ML service for natural and accurate language translation
Allows you to localize content for international users, and to easily translate large volumes of text efficiently
Transcribe
AWS ML service for automatically converting speech to text
Uses deep-learning process known as Automatic Speech Recognition to convert speech to text quickly and accurately
Supports Automatic Language Identification for multi-lingual audio; Automatically remove PII using Redaction
Use both Custom Vocabularies and Custom Language Models for the highest accuracy
Custom Vocabularies
Transcribe feature for improving accuracy where you add specific words, phrases, domain-specific terms to a custom dictionary
Good for brand names, acronyms, and other terms that Transcribe can have difficulty understanding
Increase recognition of a new word by providing hints, such as pronunciation
Custom Language Models
Transcribe feature for improving accuracy where you train the Transcribe model on your own domain-specific text data
Good for transcribing large volumes of domain-specific speech; learn the context associated with a given word
Transcribe Medical
AWS ML service for automatically convert medical-related speech to text, and is HIPAA compliant
Ability to transcribes medical terminologies such as medicines, procedures, conditions, diseases, etc.
Supports both real-time (microphone) and batch (upload files) transcriptions
Create voice apps that enable physicians to dictate medical notes; transcribe phone calls that report on drug safety and side effects
Polly
AWS ML service for turning text into lifelike speech using deep learning, allowing you to create apps that talk
Lexicons to define how to read certain specific pieces of text
Speech Synthesis Markup Language, which is Markup for your text to indicate how to pronounce it
Voice Engines, such as generative, long-form, neural, standard, and more
Speech Marks to encode where a sentence/word starts or ends in the audio; helpful for lip-syncing or highlight words as they’re spoken
Rekognition
AWS ML service for finding objects, people, text, scenes in images and videos
Facial analysis and facial search to do user verification, people counting
Create a database of “familiar faces” or compare against celebrities
Custom Labels
Rekognition feature for identifying the objects, logos, and scenes in images that are specific to your business needs
For example, you can find your logo in social media posts, identify your products on store shelves, classify machine parts in an assembly line, etc.
Label your training images and upload them to Amazon Rekognition, which will creates a custom model on your images set
New subsequent images will be categorized the custom way you have defined
Content Moderation
Rekognition feature for automatically detect inappropriate, unwanted, or offensive content
For example, filter out harmful images in social media, broadcast media, advertising, etc.
Bring down human review to 1-5% of total
content volume; integrated with Amazon Augmented AI for human review
Adaptors allow you to provide own set of labeled images; enhances the accuracy of this feature or create a specific use case of this feature
Forecast
AWS ML service for delivering highly accurate forecasts; 50% more accurate than looking at the data itself
Reduce forecasting time from months to hours
Useful for product demand planning, financial planning, resource planning, etc.
Upload data to S3; this service extracts it produces trained forecasting model for making predictions based on your needs
Lex
AWS ML service for building chatbots quickly for your apps using voice and text
Supports multiple languages; integrated with Lambda, Comprehend, Connect, Kendra, and more
Bots automatically understand the
user intent to invoke the correct Lambda function to “fulfill the intent”
Will ask for Slots, i.e. input parameters, if necessary
Personalize
AWS ML service for building apps with real-time personalized recommendations
Same technology used by Amazon; integrates into existing websites, apps, SMS, email marketing systems. etc.
Implement in days, not months; useful for retail stores, media/entertainment, etc.
Examples include personalized product recommendations/re-ranking, customized direct marketing
Textract
AWS ML service that extracts text, handwriting, and data from any scanned
documents using AI and ML
Extract data from forms and tables; read and process any type of document
Useful for Financial Services involving invoices and financial reports; Healthcare with medical records and insurance claims
Kendra
AWS ML service for fully managed document search service powered by ML
Extract answers from within a document ; text, PDF, HTML, MS PowerPoint/Word, etc.
Natural language search capabilities; learn from user interactions/feedback to promote preferred results
Ability to manually fine-tune search results by various factors like importance of data, freshness, or custom metrics
Mechanical Turk
AWS service that acts as a crowdsourcing marketplace to perform simple human
tasks
For example, you need 100k images labeled; distribute them on this marketplace and humans will tag them
You set the reward per image; useful for image classification, data collection, business processing, etc.
Integrates with Amazon A2I, SageMaker Ground Truth, and more
Augmented AI
AWS service that brings human review of ML predictions to all developers
Removes heavy lifting associated with building human review systems or managing many human reviewers
Reviewers can be your own employees, AWS contractors, MTurk workers, etc.
Your ML model can be built on AWS or elsewhere, like SageMaker, Rekognition, etc.
DeepRacer
AWS service that provides a fully autonomous race car, 1/18th in scale, driven by Reinforcement Learning
Real intent of this service is to teach ML to developers and provide guidance on building reinforcement learning models
Use the Console to train and develop RL models, backed by SageMaker, in a simulated 3D environment
Deploy RL models to Vehicle for autonomous driving; compete in the League, which hosts virtual and phyiscal events
Trainium
AWS ML computer chip built to perform Deep Learning on 100B+ parameter models
50% cost reduction when training a model, compared to comparable EC2 instances
For example, Trn1 instance has 16 Accelerators
Inferentia
AWS ML computer chip built to deliver inference at high performance and
low cost
Inf1 and Inf2 instances use this specialized chip
Up to 4x throughput and 70% cost reduction, compared to comparable EC2 instances
