Please enable JavaScript.
Coggle requires JavaScript to display documents.
ML (Amazon SageMaker, Amazon Rekognition, Amazon Comprehend, Amazon…
ML
Amazon SageMaker
Stages
2 Create an Endpoint Configuration - this let you specify the model to use, inference instance type, instance count, variant name and weight. Also called "production variant"
3 Create and Endpoint - this is where the model is published. Model is invoked with method InvokeEndpoint()
-
Elastic Inference
- It is used to reduce the cost of ML inferencies
- It speeds up throughput and decrease latency or real-time inferences deployed on SageMake hosted service using only CPU-based instance. More cost effective than a full GPU instance
- It must be configured when you create a deployable model
- Not available for all algorithms yet
Deployment Types
Offline Usage
- Asynchronous or Batch
- Generate predictions for an entire dataset all at once
- SageMaker batch transform
- Input and Output varies depending on algorothm
Online Usage
- Synchronous or Real-Time
- Generate low-latency predictions
- SageMaker hosting services
- Input varies depending on algorithm, Output is JSON string
Automatic Scaling
- Dynamically add and remove instances to a production variant based on changes in workload.
- You define and apply scaling policy that uses CloudWatch metric and target value
- Ground Truth - data labelling setup and manage jobs for training datasets using active learning and human labelling
- Notebook - access a managed Jupyter Notebook environment, here you create you models
- Processing - preprocess dataset before inference (convert format, transform feature in more expressive one, rescale/normalize, Cleaning and tokenize text, ...)
- Training - train and tune models
- Inference - package and deploy your ML models at scale
- Studio - single web-based interface for end-to-end ML development
- Data Wrangler - simplify data preparation / feature engineering, including: data selection, cleansing, exploration, visualization, and processing at scale from a single visual interface
High Availability
- AWS recommend deploying SaheMaker Container Hosts in at least 2 AZs
- Configure your CloudWatch metrics to detect when an instance goes down so that you can re-create in a new AZ
Exam Tips
- This is the service you create, train and run your ML models
SageMaker Neo
- Allows to customize you ML models for specific CPU architecture: ARM, Intel, NVIDIA
- Includes compiler to convert the ML model to an env that is optimized for the target architecture
- Support for: TensorFlow, Mxnet, PyTorch, Onnx, XGBoost
Amazon Rekognition
- Is a computer vision service that automates the recognition of pictures and videos using deep learning and neural networks
- Use this service to understand and label what is in pictures and videos
Use Cases
- Searchable image and video libraries makes images and videos searchable so you can discover objects and scenes that appear within them
- Face-based user verification enables your applications to confirm user identities by comparing their live image with a reference image
- Detection of Personal Protective Equipment such as face covers, head covers, and hand covers, glasses
- Content Moderation automatically moderate content allowing your app and websites to be family friendly
- Sentiment and demographic analysis interprets emotional expressions such as happy, sad, or surprise, and demographic info: gender from facial images
- Celebrity Recognition and label them
- Streaming Video Events Detection automatically recognizes people, animals and faces and create alert notifications (e.g. security camera)
- Custom labels identify the objects and scenes in images that are specific to your business needs
Exam Tips
- Scenario question about content moderation using AI/ML
- Make sure you app or website is family friendly
Key features:
- Custom Labels
- Faces detection and analysis (gender, age range, emotions...)
- Celebrity Recognition
- Face Search (search for a specific faces)
- People path (location and path track)
- Personal Protective Equipment (PPE) worn by persons
- Text detection
- Inappropriate or offensive content
-
Content Moderation
- Detect content that is inappropriate, unwanted, or offensive (image and videos)
- Used in social media, broadcast media, advertising, and e-commerce situations to create a safer user experience
- Set a Minimum Confidence Threshold for items that will be flagged
- Flag sensitive content for manual review in Amazon Augmented AI (A2I)
- Image --> Rekognition --> Confidence Level & Threshold --> (optional) manual review in A2I
Amazon Comprehend
Uses NPL to understand the meaning and sentiment in your text.
It is a way of automating comprehension at scale.
Use Cases:
- Analyze Call Center customers interactions (use in conjuntion with Amazon Transcribe)
- Index & Search Product Reviews
- Legal Briefs Management to search contracts and examples of recent court cases
- Process Financial Documents e.g. insurance companies can automate claims for common cases
-
Key Features:
- Language of the text
- Extracts key phrases, places, people, brands, or events
- Understands how positive or negative the text is
- Analyzes text using tokenization and parts of speech
- Automatically organizes a collection of text files by topic
Amazon Textract
- Use ML + OCR to automatically extract text, handwriting and data from scanned documents and images
- Goes beyond OCR with the help of ML (OCR + ML) to extract text without human intervention
- Financial Information
- Health Care and Life Sciences
- Public Sector
Key features:
- Detect typed and handwritten text in a variety of documents, including financial reports, medical records, and tax forms
- Extract text, forms, and tables from documents with structured data using the Amazon Textract Document Analysis API
- Specify and extract information from documents using the Queries feature within the Amazon Textract Analyze Document API
- Process invoices and receipts with the AnalyzeExpense API
- Process ID documents, such as driver’s licenses and passports issued by the U.S. government, using the AnalyzeID API
- Upload and process mortgage loan packages through automatic routing of the document pages to the appropriate Textract analysis operations using the Analyze Lending workflow
Amazon Forecast
-
You send your data to Amazon forecast and it will automatically learn your data, select the right algorithm and then forecast your data
-
Amazon Lex
-
Use Cases:
- Virtual Agent
- Voice Assistant
- Automate Informational Responses (automate FAQ)
- Improve Productivity with chatbots
- Automate Contact Center scripts as chatbot
Key features:
- Automatic Speech Recognition (ASR) to convert speech to text
- Natural Language Understanding to recognize the intent of text, caller
- Helps build chatbots, call center bots
Amazon Polly
Turns your text into lifelike speech and allows you to create applications that talk to and interact with using a variety of languages and accents
Use Cases:
- Content Creation
- Make content more accessible
- Convert a blog into lifelike speech
Features
Speech Marks are metadata that describe the speech that you synthesize, such as where a sentence or word starts and ends in the audio stream. Use Cases: combining the metadata with the audio stream from your text can enable you to synchronize speech with facial animation (lip-syncing) or to highlight written words as they're spoken
- Speech Synthesis Markup Language (SSML) generate speech from either plain text or from documents marked up with Using SSML-enhanced text that enables more customization
- emphasizing specific words or phrases
- using phonetic pronunciation
- including breathing sounds, whispering
- using the Newscaster speaking style
- long pause
- speech rate
Neural TTS (NTTS) system that can produce even higher quality voices than its standard voices. The NTTS system produces the most natural and human-like text-to-speech voices possible
- Pronunciation Lexicons enable to customize the pronunciation of words
- Use Cases:
- Acronyms:AWS=>“AmazonWebServices”
- Stylized words: St3ph4ne => “Stephane”
- words uncommon to the selected language
- Upload the lexicons and use them in the SynthesizeSpeech operation
Amazon Kendra
Uses ML to provide a centralized search service across different silos of unstructured data (S3, file servers, websites ...)
Use Cases:
- Accelerate R&D searching papers all over the places
- Improve Customer Interactions better understanding customer's asks and returning relevant answers
- Minimize Regulatory and Compliance risks by researching new/updated regulatory that may impact you business
- Increase Employee Productivity by having your data searchable in a single place
Amazon Fraud Detector
Analyze your data, train a model to detect frauds in your data and then actively searches for frauds
- Identify suspicious Online Payments based on previous cases of fraud
- Detect New Account Fraud creating a ML model to distinguish between genuine new accounts/users and high-risk accounts
- Prevent Trial and Loyalty Program Abuse stops users from automation "free-trial" accounts
- Improve Account Takeover Detection identify where an illegitimate user hijacks a legitimate user's account
Amazon Translate
A deep-learning and neural-network ML service allowing you to translate from one language to another
Use Cases:
- Highly Accurate and Continuously Improving
- Easy to integrate with your App
- Cost Effective when compared with the cost of hiring a translator
- Scalable
Amazon Transcribe
- Converts speech to text automatically
- Automatically remove Personally Identifiable Information (PII) using Redaction
- Supports Automatic Language Identification for multi-lingual audio
- Use Cases:
- Convert Audio and Video Files into Text
- Generate subtitles on the fly
- Generate metadata for media assets to create a fully searchable archive
Amazon Personalize
Key Features
- Fully managed ML-service to build apps with real-time personalized recommendations
- Same technology used by Amazon.com
- Integrates into existing websites, applications, SMS, email marketing systems, ...
- Implement in days, not months (you don’t need to build, train, and deploy ML solutions)
- Import data from S3 or Personalize API
Use cases
- Retail stores
- Media
- Entertainment
-