MachineLearning
Comprehend
- natural language processing service
- input = document
- output = entities, key phrases, language, sentiment, PII ...
- pre-trained models or custom
real-time
analysis for small workloads
async
jobs for larger workloads
Forecast
- forecasting for time-series data
- import historical and related data
- output = forecast and forecast expandability
Fraud Detector
- fully managed fraud detection service
- upload historical data and choose model type
- model type
online fraud
: little historical data
transaction fraud
: suspect payments
account takeover
: phishing or other social based attack
Kendra
intelligent search service
- supports wide range of question types
- types of questions
factoid
: who, what, where
descriptive
: how
keyword
index
: searchable data organized in an efficient way
data source
: where data lives
- kendra connects and indexes from this location
- synchronize with index on a schedule
- documents :
structured or unstructured
Lex
text or voice conversational interfaces
- automatic
speech recognition
(ASR)
- natural
language understanding
(NLU)
bots
- converse in 1+ languages
- utterances : different ways something can be said
- how to fulfil intent
- slots
- parameters for intent
Polly
- turns text into lifelike speech
- tts types
- standard tts
- neural tts (much better)
- output formats : mp3, ogg vorbis ...
- Speech Synthesis Markup Language (SSML)
- addition control on how polly generates speech
- emphasis, pronunciation, whispering ...
Rekognition
- deep learning based image and video analysis
- identify object, people, content moderation ...
- pay per image
SageMaker
- fully managed machine learning service
- fetch, clean, prepare, train, evaluate, deploy, monitor
- sagemaker studio : IDE for ML lifecycle
- sagemaker domain : isolation/grouping for a project
- containers : docker containers deployed to ml ec2 instance
- hosting : deploy endpoints for your models
- no cost : but the resources it creates do
- detect and analyze text contained in input documents
- input = JPEG, PNG, PDF, TIFF
- synchronous and asynchronous
- pay for usage
- detection of text
- relationships in text, metadata
- extracts fields from documents, receipts, IDs
Transcribe
- automatic speech recognition (ASR)
- input = audio
- output = text
- pay as you use
- options
- language customizable, filers for privacy, audience appropriate language, speaker identifier, custom vocabulary
- use case
- index audio to allow searching
- meeting notes
- subtitles/captions and transcripts
- call analytics
Translate
- text translation service
- translates text from one language to another
- ensures meaning is translated
- autodetect source text language
- use case
- multi-language user experience