Skip to content

AWS Machine Learning

Amazon Comprehend

  • Natural Language Processing (NLP)
  • Input = Document
  • Output = Entities, phrases, language, PII, sentiments
  • Pre-trained models or custom
  • Real-time analysis for small workloads
  • Async jobs for larger workloads
  • Console, CLI for interactive or use APIs to build into applications

Amazon Kendra

  • Intelligent search service designed to mimic interacting with human expert
  • Supports wide range of question types
  • Factoid: Who, What & Where
  • Descriptive: How do I ..?
  • Keyword: What day is the keynote address?
  • Index: searchable data organized in an efficient way
  • Data source: Where your data lives. Kendra connects and indexes from this location
  • Synchronize with index based on a schedule
  • Integrates with AWS Services (IAM, Identity Center)

Amazon Lex

  • Text or Voice consersational interfaces
  • Powers the Alexa service
  • Automatic speech recognition (ASR) - speech to text
  • Natural Language Understanding (NLU) - Intent
  • Build understanding into you application
  • Scales, integrates, Quick to deploy, pay as you go pricing
  • Chatbots, voice assistants, Q&A bots, Info/Enterprise Bots
  • Lex can fulfill the intent using lambda integration

Amazon Polly

  • Converts text into "life-like" speech
  • Text (language) => Speech (language) NO TRANSLATION
  • Modes: Standard TTS
  • Modes: Neural TTS
  • Can use Speech Synthesis Markup Language (SSML): Additional control over how polly generates speech (emphasis, pronunciation, whispering)

Amazon Rekognition

  • Deep learning image & video analysis
  • Identify object, people, text, activities, content moderation, face detection, face analysis, face comparison, pathing & much more
  • Pay per image or per minute video
  • Integrates with application & event driven
  • Can analyse live video streams - kinesis video streams

Amazon Textract

  • Detect and Analyse text contained in input documents
  • Input = JPEG, PNG, PDF or TIFF
  • Output = Extracted text, structure & Analysis
  • Synchronous (real-time) for small documents and Asynchronous for large files
  • Pay per usage. Custom pricing available for large volume
  • Use case:
    • Detection of text & relationship between text
    • generate metadata
    • Document analsysis (names,addresses, birthdate)
    • Receipt analysis (prices, vendor, line-items, dates)
    • Identity documents (document ID)

Amazon Transcribe

  • Automatic Speech Recognition (ASR) service
  • Input=Audio, Output=Text
  • Language customization, filters for privacy, audience-appropriation language, speaker identification
  • Custom vocabularies and language models
  • Pay as you use per second of transcribed audio.
  • Use case:
    • Full text indexing of audio - allow searching
    • Meeting notes
    • Subtitles/Captions & transcripts
    • Call analytics(characteristics, summarization, categories and sentiment)
    • Integration wiht other apps / AWS ML services

Amazon Translate

  • Text translation service based on ML
  • Translates text from native language to other languages. One word at a time.
  • Encoder reads source => semantic representation (meaning)
  • Decoder reads meaning => writes target language
  • Attention mechanism ensures 'meaning' is translated
  • Auto detect source text language
  • Use case:
    • Multilingual user experience
    • Translate incoming data (social media, news, communications)
    • Language independence for other AWS services (Comprehend, transcribe, polly, data stored in S3, RDS, Database)

Amazon Forecast

  • Forecasting for time-series data
  • Retail demand, supply chain, staffing, energy, server capacity, web traffic
  • Import historical & related data, Output forecast and forecast explainability
  • Web Console (visualization), CLI, APIs, Python SDK

Amazon Fraud Detector

  • Fully managed Fraud detection service
  • New account creations, payments, guest checkout
  • Upload historical data, choose model type
  • Model Types:
    • Online Fraud - Little historical data eg: new customer account
    • Transaction Fraud - transactional history, idenitfying suspect payments
    • Acount Takeover - Identify phishing or another social based attack
  • All events are scored. Rules/Detection logic allow you to react to a score based on business activity

Amazon SageMaker

  • Fully managed Machine Learning (ML) service
  • Fetch, Clean, Prepare, Train, Evaluate, Deploy, Monitor/Collect
  • Sage Maker Studio: Build, train, debug and monitor models - IDE for ML lifecycle
  • Sagemaker Domain - EFS Volume, Users, Apps, Policies, VPCs -- isolation
  • Container - Docker containers deployed to ML EC2 instance - ML environments (OS, libraries, tooling)
  • Hosting - Deploy endpoints for your models
  • Sagemaker has no cost - the resources it creates have a cost (Complex pricing)