Skip to main content

Machine Learning

Machine Learning​

Rekognition​

Helps us to do segmentation, object detection, and OCR in images and videos

In general helps us to place bounding boxes and label them

Use Cases:

  • Labeling
  • Content moderation
    • Helpful in detecting harmful cntent which is unwanted or offensive
    • Returns a content threshold with a confidence level
    • This allows us to do Human manual review via Amazon Augmented AI (A2I)
  • Text detection
  • Face detection and analysis
  • Face search and verification
  • Celebrity recognition
  • Pathing (like sports game analysis)

Transcribe​

Automatically convert speech to text

Uses Automatic Speeech Recognition (ASR), and lets us automatically remove PII with automatic language identification

Use cases:

  • Customer service calls
  • Automated closed captioning
  • Generating metadata for media assets for archiving and search

Polly​

Does the opposite of Transcribe!

Text to Speech. Allows us to create applications that can talk

  • Lexicons help us to altar what text should be read as compared to what it's typed as

  • Speech markup language allows us to style the kind of speaking that comes out

  • Customize pronunciation of words with Pronunciation Lexicons

    • Stylized words "L337" should be pronounced "Leyt"
    • "AWS" -> Amazon Web Services
    • etc
    • Upload them and use in the SynthesizeSpeech operation

  • Generate speech frmo plain text or from documents marked up with Speech Synthesis Markup Language (SSML)

    • Emphasizing specific words or phrases!
    • Using phönetic pronunciation
    • Including breathing sounds, or whispering
    • Using the Newscaster speaking style

Translate​

Allows us to do natural and accurate language translation

  • Localize content such as websites and applications
  • Also does for international users, and allows us to easily translate large volumes of text efficiently

Lex​

Same thing that powers Amazon Alexa - helps to understand intent

Automatic speech recognition that allows us to convert speech into text

  • Automated Speech Recognition to convert speech to text
  • Natural language understanding to recognize intent of text, callers
  • Integrates with Amazon Connect
    • Amazon connect is for cloud-based virtual contact center
    • Connect will accept calls and stream into Lex, it will understand intent and invoke lambda, this will schedule something in CRM
    • Receive calls, create contact flows, and help establish cloud based customer contact center
    • Can integrate with other CRM systems or AWS Services
    • No upfront payments, and 80% cheaper than traditional contact center solutions

The entire solution of Lex is around automating speech understanding in scenarios, most specifically contact centers or customer service

Comprehend​

  • NLP!
  • Fully managed and serverless
  • Use cases:
    • Language detection
    • Key phrases, places, people, brands, or events
    • Sentiment analysis
    • Analyzes text using tokenization and parts of speceh
    • Organizes and collection of text files by topic
      • Topic modeling

Comprehend Medical​

  • Detects and returns useful information in unstructured clinical text
    • Physicians notes
    • Dischare summaries
    • Test results
    • etc
  • Uses NLP to detect Protected Health Information (PHI)
  • Upload all documents into S3 and have Comprehend Medical analyze it

Sagemaker​

Fully managed service for developers / data scientists to build ML models

  • Helps to create ML Platforms
  • Typically difficult to do all processes of typical ML platform in one place while provisioning servers - all of these services are covered:
    • Data
      • Labeling
      • Transformations
      • Feature creation
    • Modeling
      • Training
      • Tuning
    • Deployment
      • New data coming in
      • Feedback loop
      • Inference code

Kendra​

Document Search Service

  • Helps us to extract answers from within a document
  • Natural language search abilities
    • "Where are the offices" - indexes all of our documents and returns an answers
  • Able to manually finetune results

Personalize​

  • Fully managed ML service to build apps with real-time personalized recommendations
  • Same product used by amazon.com
  • Read input data from S3, or Amazon Personalize API for real time
  • Integrates into existing websites, apps, SMS, email marketing systems, etc
  • Implements in days, not months (no need to build, train, and deploy)
  • Use cases:
    • Retail store shopping page
    • Media and entertainment landing page
    • etc

Textract​

Extract text, handwriting, and other data from any scanned documents using AI and ML

  • Results given back as data.json file
  • Read and process any type of document (PDF, images, etc)

Alexa For Business​

  • Use ALexa to help employees be more productive in meeting rooms and their desks
  • Measure and increase utilization of meeting rooms in workspace