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)
AI - Software that imitatates human behavior and capabilities
- Machine Learning
- Anomaly Detection
- Computer Vision
- Natural Language Processing
- Knowledge Mining
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
Analyses data overtime and detect unusual changes
- Anomaly Detection service API
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
AI dealing with undesrtanding written and spoken language
Natural language processing (NLP) in Microsoft Azure
- Language
- Translator
- Speech
- Azure Bot
Extracting information from large amount of unstructured data
Knowledge mining in Microsoft Azure
- Azure Cognitive Search
- Bias results
- Errors causing harm
- Data exposure
- Not compatible for everyone
- User's need to trust complexity
- Liability issues for decisions
- Fairness - not biased for anyone
- Reliability and safety
- Privacy and Security
- Inclusiveness - must empower everyone
- Transparency - user's should be aware
- Accountability - governance structure
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
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
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
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