Introduction to AWS Machine Learning AI Services for Pythonistas

Piotr Grzesik

What I do ?

What I will be talking about ?

• What are the benefits of Machine Learning APIs ?
• Overview of ML APIs services on AWS
• How to use Machine Learning APIs in Python apps ?

Building your own solution

Using Machine Learning APIs

Benefits of ML APIs

• Does not require specific ML knowledge (can be used by "regular" developers)
• Easy to consume via API/SDK
• Trained on a large dataset
• Models are constantly re-trained/improved
• "Infinitely" scalable
• Pay-per-use

Cons of ML APIs

• Limited functionality, suitable only for common tasks 
• Can be much pricier than custom solution in the long run
• No way to customize/tweak models (there are small exceptions)
• Vendor lock-in (not really)

Providers

What AWS has to offer ?

• Rekognition Image/Video
• Polly
• Comprehend
• Transcribe
• Translate
• Textract
• Lex

Rekognition

• Image and video recognition service
• Supports processing images from S3 buckets
• Supports batch and real-time processing
• 1000 minutes of video analysis and 5000 images per month available in Free Tier
• Alternatives: Google Cloud Vision API, Azure Computer Vision API

Rekognition Image features

• Object and scenery detection
• Facial recognition
• Facial analysis
• Face comparision
• Inappropriate image detection
• Celebrity recognition
• Text in image

Rekognition Video features

• Person identification, tracking and pathing
• Face recognition
• Facial analysis
• Objects, scenes and activities detection
• Inappropriate video detection
• Celebrity recognition

Rekognition Image demo

Rekognition use cases

• Detect unsafe/inappropriate images and videos
• Indexing images/videos based on scenery and/or emotions
• Blurring faces on images and videos
• User verification based on face image
• License plate verification
• Reading handwritten text from images

Polly

• Speech synthesis service
• Supports near real-time and batch processing
• 5 million characters for speech or Speech Marks requests per month available in Free Tier
• Alternatives: Google Cloud Text-To-Speech

Polly features

• Supports mp3, Vorbis and PCM formats
• Supports generating "Speech Marks"
• Supports SSML (Speech Synthesis Markup Language)
• Supports PLS (Pronounciation Lexicon Specification)

Polly demo

Polly use cases

• Wordpress Polly plugin
• Combine Rekognition and Polly to provide audio files from handwritten notes
• Language learning applications
• Voicing dialogs in games
• Creating audiobooks

Comprehend

• Natural language processing (NLP) service
• Integrates with Amazon S3 and AWS Glue
• Supports batch processing
• Supports English, Spanish, French, German, Italian, Portuguese
• 5 million characters per month available in Free Tier
• Alternatives: Google Cloud Natural Language API, Azure Language API

Comprehend features

• Keyphrase extraction
• Sentiment analysis
• Syntax analysis
• Entity recognition
• Language detection
• Topic modeling
• Medical Named Entity extraction
• Custom classification models

Comprehend demo

Comprehend use cases

• Analyze sentiment in hotel reviews
• Generate tags for articles
• Group documents by topic
• Search text based on key phrases or sentiment

Transcribe

• Automatic speech recognition service
• Supports Spanish, English, French, Portuguese, Hindi, Korean, German
• Supports multiple speakers and custom vocabulary
• Optimized for Telephony Audio
• Supports channel indentification
• 60 minutes per month available in Free Tier
• Alternatives: Google Cloud Speech API, Azure Speech API

Transcribe demo

Transcribe use cases

• Generate podcast transcripts
• Generate notes from board meetings
• Combine with Comprehend for sentiment analysis of speech
• Generate transcripts of telephone calls
• Combine with Translate for generation of translated movie subtitles

Translate

• Text translation service
• Supports real-time and batch processing
• Supports Named Entity Translation Customization
• Automatically detects source language
• Supports 25 languages, 595 translation combinations 
• 2 million characters per month available in Free Tier
• Alternatives: Google Cloud Translation API, Azure Translator Text API

Translate use cases

• Automatic translation of customer reviews
• Automatic translation of company website
• Automatic translation of emails
• Translation support in conversational applications
• Translation capabilities in language learning applications
• Integrate with Polly to provide audio files in multiple languages

Textract

• Document analysis service, extracts text and data
• Supports PDF, PNG and JPG
• Currently only supports English
• Support for form and table extraction
• 1000 pages per month for Detect Document Text API and 100 pages per month for Analyse Document API available in Free Tier
• Alternatives: Google Cloud Cloud Vision API, Microsoft Azure Form Recogniser API [preview]

Textract use cases

• Extract text that can be later analyzed with Comprehend
• Automatically extract key data from paper forms and documents
• Combine it with Polly to provide audio files from processed documents

Lex

• Service for building conversational interfaces
• Supports both voice and text inputs
• Uses the same technology as Amazon Alexa
• 10000 text requests and 5000 voice requests per month available in Free Tier
• Alternatives: Dialogflow, Azure Bot Service

Lex features

• Speech recognition
• Natural language understanding
• Multi-turn conversations
• Confirmation and error prompts
• Intent chaining

Lex use cases

• Customer support Messenger bots
• Learning bots that asks questions and verify answers
• Bots for adding voice control to connected devices e.g. smart bulbs

Summary

If you want to...

• analyze images and videos, use Rekognition
• generate audio from text, use Polly
• analyze text, use Comprehend
• recognize speech, use Transcribe
• translate text, use Translate
• extract text from documents, use Textract
• build a chatbot, use Lex

Thanks!

@p_grzesik

pj.grzesik@gmail.com