Skip to content

Latest commit

 

History

History
351 lines (219 loc) · 10.5 KB

README.md

File metadata and controls

351 lines (219 loc) · 10.5 KB

AI-900 Study Notes

AI-900 Exam guide: https://docs.microsoft.com/en-gb/learn/certifications/exams/ai-900

LEARNING PATH 1: Get started with artificial intelligence (AI-900)

LEARNING PATH 2: Explore visual tools for machine learning (AI-900)

LEARNING PATH 3: Explore computer vision (AI-900)


LEARNING PATH 1: Get started with artificial intelligence (AI-900)

What is AI?

AI - Software that imitatates human behavior and capabilities

  • Machine Learning
  • Anomaly Detection
  • Computer Vision
  • Natural Language Processing
  • Knowledge Mining

Understand machine learning

Use massive amount of data for training machine learning models that can predict future outcomes

Machine learning in Microsoft Azure

  • Automated Machine learning
  • Azure Machine learning designer
  • Data and compute management
  • Pipelines

Understanding anomaly detection

Analyses data overtime and detect unusual changes

  • Anomaly Detection service API

Understand computer vision

AI dealing with Visual Processing

Common Computer Vision tasks,

  • Image Classification
  • Object Detection
  • Semantic Segmentation
  • Image Analysis
  • Face Detection
  • OCR - Optical Character Recognition

Computer Vision in Microsoft Azure

  • Computer Vision
  • Custom Vision
  • Face
  • Form Recognition

Understand natural language processing

AI dealing with undesrtanding written and spoken language

Natural language processing (NLP) in Microsoft Azure

  • Language
  • Translator
  • Speech
  • Azure Bot

Understand knowledge mining

Extracting information from large amount of unstructured data

Knowledge mining in Microsoft Azure

  • Azure Cognitive Search

Challenges and risks with AI

  • Bias results
  • Errors causing harm
  • Data exposure
  • Not compatible for everyone
  • User's need to trust complexity
  • Liability issues for decisions

Understand Responsible AI

  • Fairness - not biased for anyone
  • Reliability and safety
  • Privacy and Security
  • Inclusiveness - must empower everyone
  • Transparency - user's should be aware
  • Accountability - governance structure

LEARNING PATH 2: Explore visual tools for machine learning (AI-900)

Machine Learning

Uses mathematics and statistics to create a model that can predict future unknown values

Supervised machine learning approach

  • Regression: predict a continuous value
  • Classification: determine a binary class label

Unsupervised machine learning approach

  • Clustering: used to determine labels by grouping similar information into label groups

Azure Machine Learning studio

Train and deploy effective with a lot of simplicity and automation

You must create Workspace in your Azure subscription first

Computer Resources you can create,

  • Computer Instances
  • Computer Clusters
  • Inference Clusters
  • Attached Compute

Azure Automated Machine Learning

  • Automated machine learning capability that automatically tries multiple pre-processing techniques and model-training algorithms in parallel

...

WIP

LEARNING PATH 3: Explore computer vision (AI-900)

Azure resources for Computer Vision

  • Computer Vision
  • Cognitive Services (includes Computer Vision and other capabilities ex: Text & Language)

Computer Vision capabilities

  • Describe image
  • Tagging Visual features
  • Detecting Objects
  • Detecting Brands
  • Detecting Faces
  • Categorizing an image
  • Detect domain specific content
  • OCR
  • Generate thumbnails, moderate content, etc

Uses of image classification

  • Product identification
  • Disaster investigation
  • Medical diagnosis

Classification - predict which category or class something belogs to. Image classification is a machine learning technique.

Convolutional neural networks (CNNs) - to uncover patterns in the pixels that correspond to particular classes

Azure resources for Custom Vision

  • Custom Vision - dedicated resource can be training, a prediction, or both
  • Cognitive Services - eneral cognitive services resource, includes Custom Vision with many other

It is possble to mix and match, ex: use a dedicated Custom Vision resource for training, but deploy your model to a Cognitive Services resource for prediction. This can only be done in same region!

Model evaluation

  • Precision - the model predicted 10 images are oranges, but only 8 were actually oranges, then the precision is 0.8 (80%)
  • Recall - the model prediction 7 images are apples, but there are 10 images of apples, then the recall is 0.7
  • AP - Average prediction combination of previous two

Uses of object detection

  • Detecting tumors
  • Driver assistance
  • Checking for building safety

Use Custom Vision or Congnitive Services resource, but better with a mix

To train an object detection model, you need to create a Custom Vision project based on your training resource

Steps:

  • Image tagging - Custom Vision portal provides a graphical interface, for training data set up
  • Model training and evaluation
  • Using the model for prediction

Face detection

  • Facial analysis - information on facial features
  • Facial recognition - indentify a paticular face

Microsoft Azure provides multiple cognitive services,

  • Computer Vision - basic face detection
  • Video Indexer - Detect faces in a video
  • Face - Blur, Exposure, Glasses, Head Pose, Noise, Occlusion

Following are restricted and customers should fill intake form,

  • The ability to compare faces for similarity
  • The ability to identify named individuals in an image

Use the Computer Vision service to read text

The Read API - scanned large documents

  • This is an asynchronous service
  • Structure returned - Pages, Lines, Words

Process receipt or invoices data

Azure Form Recognizer

Two options,

  • Pre built receipt model
  • Custom model - trained with your data

LEARNING PATH 3: Explore natural language processing (AI-900)

Evaluate different aspects of a document or phrase, in order to gain insights into the content of that text

In Microsoft Azure, the Language cognitive service

  • Determine language of a document
  • Perform sentiment analysis
  • Extract key phrases
  • Identify and categorize entities

Two main capabilities:

  • Speech recognition - acoustic model -> language model -> to text
  • Speech synthesis - tokenizes text -> prosodic units (such as phrases, clauses, or sentences)

Azure resources for the Speech service

  • Speech resource - only for speech service
  • Cognitive Services resource - including other services as well

Available APIs,

  • Speech-To-Text-API - Real time transcription, batch transcription features
  • Text-To-Speech-API - Speech synthesis voices features

Two types of translations

  • Literal translation - word to word translation
  • Semantic translation - translation with context

Azure Cognitive Services provides,

  • Translator Service - Text to Text translation
  • Speech Service - Speech to Text / Speech to Speech translation

The service uses a Neural Machine Translation (NMT) model for translation, which analyzes the semantic context of the text

Use ISO 639-1 language codes, such as 'en' for English Use ISO 3166-1 cultural code - for example, en-US for US English

Optional configuration,

  • Profanity filtering
  • Selective translation (mark some parts not to translate)

The Speech service APIs,

  • Speech-to-text - used to transcribe speech from an audio source to text format.
  • Text-to-speech - used to generate spoken audio from a text source.
  • Speech Translation - used to translate speech in one language to text or speech in another.

You can specify a single "from" language and multiple "to" languages.

Microsoft Azure, conversational language understanding is supported through the Language Service

  • Utterances - something a user might say
  • Entities - item to which an utternce refers
  • Intents - represents a purpose or goal

The 'None' intent is used to handle fallback/default responses, its mandatory, cannot remove or edit

Authoring the model - First you must define entities, intents, and utterances with which to train the language model

Types of Entities,

  • Machine Learned
  • List
  • Reg-ex
  • Pattern.any

Connect clients using prediction resource and authentication key

Microsoft Azure provides two core services for creating bot solutions,

  • Language Service
  • Azure Bot Service

First provision a Language service resource in Azure to create a 'Knowledge base' Use the Language Studio's custom question answering feature to create, train, publish, and manage knowledge bases

Question-and-answer pairs are created by,

  • Generated from an existing FAQ page
  • Entered and edited manually
  • Mix of both above

The client applications require:

  • The knowledge base ID
  • The knowledge base endpoint
  • The knowledge base authorization key

Create a Bot with Azure Bot Service - easier to use automatic bot creation feature

Connect multiple channels to the same bot for responding to users

...

WIP


That's all the notes, good luck with the AI-900 exam! ^_^

-Udara Alwis