AI-900: Microsoft Azure AI Fundamentals Exam: Quick Revision Notes

1. Supervised Learning: Find patterns in data based on labeled input data. Example: Regression & Classification.
2. Unsupervised Learning: Pattern discovery without labeled data using Unsupervised Learning. Example: Clustering

A ML-based technique that analyzes data over time and identifies unusual changes.

• Credit Card transactions fraud detection.
• Network intrusion detection.
• Machine failure prediction.

Azure Service: Anomaly Detector: Foresee problems before they occur

• Image classification: Training ML model to classify images based on their contents. Example: classify images based on the type of vehicle they contain, such as taxis, buses, cyclists.
  
• Object detection: Training ML models to classify individual objects within an image, and identify their location with a bounding box.
  
• Semantic segmentation: ML technique in which individual pixels in the image are classified according to the object to which they belong.
  
• Image analysis: A combination of ML models and advanced image analysis techniques to extract information from images, including tags, descriptive captions that summarize the scene shown in the image.
  
• Face detection, analysis, and recognition: A specialized form of object detection that locates human faces in an image.
  
• Optical character recognition (OCR): Detect and read text in images
  

• Computer Vision
• Custom Vision
• Face
• Form Recognizer

• Automatically translate spoken or written phrases between languages.
• Analyze and interpret text in documents, email messages, and other sources.
• Interpret commands and determine appropriate actions.
• Interpret spoken language, and synthesize speech responses.

• Language
• Translator
• Speech
• Azure Bot

• Bias can affect results - dealing with fairness.
• Errors may cause harm - dealing with reliability.
• Data could be exposed - dealing with privacy and safety.
• Solutions may not work for everyone - dealing with inclusiveness.
• Users must trust a complex system - dealing with transparency.
• Who's liable for AI-driven decisions? - dealing with accountability.

1. Fairness: A loan-approval model discriminates by gender due to bias in the data with which it was trained
2. Reliability: An autonomous vehicle experiences a system failure and causes a collision
3. Privacy & Safety: A medical diagnostic bot is trained using sensitive patient data, which is stored insecurely
4. Inclusiveness: A home automation assistant provides no audio output for visually impaired users
5. Transparency: An AI-based financial tool makes investment recommendations - what are they based on?
6. Accountability: An innocent person is convicted of a crime based on evidence from facial recognition – who's responsible?

• Predict house rent (label) based on features: size, city, population, etc..
• Predict crop yield (label) based on features: weather, solid, temperature, etc.
• Predict future advertisement click based on past data.

• Will it snow tomorrow?
• Will he buy a car?
• Will they win the game?
• Is this Email a spam?

1. Identify distinct customer segments based on their purchase history.
2. Cluster website visitors based on their browsing interests.
3. Group social media posts based on the topics.
4. Classify patient health records to identify distinct groups with similar medical conditions
Exam Tip: If the question is related to segmenting, clustering, classification or grouping and the data is not labeled (i.e unsupervised) then the question is about clustering.

• Predict fair of a taxi-journey.
• Features: distance for the journey, traffic, type of taxi.
• Lable: fair of the taxi-journey.
• Identify label and features for sales prediction - date, time, footfalls, sales, discounts?
• Features:: date, time, footfalls, discounts
• Label: sales
• Predicting the price of a house:
• Features: Number of bedrooms, Square footage, Location, Age of the house
• Label: Price of the house
• Predicting student performance:
• Features: Study hours, Previous exam scores, Attendance, Study habits
• Label: Exam score

1. Inject Dataset
2. Prepare Dataset - Identify Features and Label, Clean and Transform the data.
3. Split the data randomly into training and validation data.
4. Train the model using ML Technique.
5. Evaluate the model with the validation data.
6. Deploy the model

It is used to assess the quality of a Classification Model's predictions.

• True Positive: Model predicts the patient has Covid-19, and the patient does actually have Covid-19.
• False Positive: Model predicts the patient has Covid-19, but the patient doesn't actually have Covid-19.
• False Negative: Model predicts the patient doesn't have Covid-19, but the patient actually does have Covid-19.
• True Negative: Model predicts the patient doesn't have Covid-19, and the patient actually doesn't have Covid-19.

• Accuracy
• Precision
• Recall (True positive rate)
• F1 Score

Used to automate your model into a service that makes continuous predictions.

Once the pipeline is ready you can do deployment of your model at an endpoint

• Prepare data
• Train a model
• Deploy a model

• Compute Instances: These are development workstations that data scientists can use to work with data and models.
• Compute Clusters: These are scalable clusters of virtual machines for on-demand processing of experiment code.
• Inference Clusters: These are deployment targets for predictive services that use your trained models.
• Attached Compute: These links are to existing Azure compute resources, such as Virtual Machines or Azure Databricks clusters.

• Content Organization
• Text Extraction
• Spatial Analysis

1. Computer Vision:

   A specific resource for the Computer Vision service.

   Exam Tip: Use this type of resource if,
   
   

   1. You don't intend to use any other cognitive services.

   2. You want to track utilization and costs for your Computer Vision resource separately.



2. Cognitive Services:

   A general cognitive services resource that includes Computer Vision along with many other cognitive services (such as Text Analytics, Translator Text, etc.)

   Exam Tip: Use this type of resource if,
   
   

   1. You intend to use multiple other cognitive services.

   2. To simplify administration and development.

   Exam Tip:

   For both the resources you get,

   1. A key: It is used to authenticate client applications.

   2. An endpoint: To provide the HTTP address at which your resource can be accessed.

1. Describing an image: A large building in a city.
2. Tagging images: Metadata that summarizes attributes of the image: example - car, tower, building.
3. Object detection: Tag with a bounding box coordinates.
4. Brands detection: Identify know brands from a image.
5. Image detection: Can detect and analyze human faces in an image, including the ability to determine age and a bounding box rectangle for the location of the face(s).
6. Image categorization: Example - people_, people_group, animal_ , animal_birds. There are 86 categories.
7. Domain-Specific Content detection: Known Celebrities and Landmarks.
8. Optical character recognition (OCR): Detect printed and handwritten text in images.
9. Image types detection: for example, identifying clip art images or line drawings.
10. Image color schemes detection- specifically, identifying the dominant foreground, background, and overall colors in an image.
11. Generate thumbnails: creating small versions of images.
12. Moderate content: detecting images that contains that violates security and privacy.

1. Text mining
2. Text analysis
3. Sentiment analysis
4. Key phrase extraction
5. Named entity recognition
6. Language detection

1. Statistical analysis: Removal of stop words (a, an..)
2. Frequency analysis: counting how often each word appears
3. N-grams: a two-word phrase is a bi-gram, a three-word phrase is a tri-gram, and so on
4. Stemming/Lemmatization: Algorithms to normalize words before counting them: example: walk, walked, walking.
5. Noun phrase: Breaking down sentences into tree-like structures such
6. Sentiment Analysis: Encoding words as numeric features to train a machine learning model. Workload: Classify a document as positive or negative sentiments.
7. Vectorized Models: Models that capture semantic relationships between words by assigning them to locations in n-dimensional space.

• Identify the language of the text - e.g. English, German, Swedish etc.
• Sentement Analysis of the text - positive/negative sentiments.
• Key phrases extraction from text for point taking.
• Identify and categorize entities in the text - people, places, organizations, dates, times, quantities etc.

• Provides the language name: example: "English"
• Provides the ISO 6391 language code: example en
• Provides a numeric score (0.1-1.0) - the confidence in the language detection

Evaluating the text of a document, or documents, and then identifying the main talking points of the document(s)

Text: Code2care is an initiative to publish and share varied knowledge in technical and technical areas gathered during day-to-day learnings and development activities. People can leverage this portal to find solutions to their various queries without re-inventing the wheel by referring to our easy to understand posts.

Key phrases:

• Code2care - tech knowledge initiative
• Varied technical knowledge sharing
• Day-to-day learnings and development
• Solutions for queries
• Easy-to-understand posts
• People can leverage this portal

From the provided unstructured text get a list of entities in the text that it recognizes.

Entity: Andrew
Type: Person
Entity: NYC Central Park
Type: Location
Entity: evening
Type: DateTime or Time

• To providing closed captions for recorded or live videos
• For creating a transcript of a phone call or meeting
• Automated note dictation
• To determining intended user input for further processing

Speech-to-text API can be used for real-time and batch transcription

To generate spoken output.

Azure Service: Speech to text API

To generate spoken output (text-to-speech)

Azure Service: Text to speech API

• To generating spoken responses to user input.
• To create voice menus for telephone systems.
• For reading email or text messages aloud in hands-free scenarios.
• To broadcasting announcements in public locations, such as railway stations or airports.

Text-to-text translation.

Uses a Neural Machine Translation (NMT) model for translation.

Supports more than 60 languages.

Can simultaneously translate a source document into multiple languages.

1. Profanity filtering
2. Selective translation: Tag code, a brand name, or a word/phrase

Speech-to-text and speech-to-speech translation.

Speech to text

Text to speech

Speech Translation - used to translate speech in one language to text or speech in another.

• Utterances: something a user might say

  Example:
  
  "Turn off the light"
  "Switch on the TV"

• Entities: Ttem to which an utterance refers.

  Example:
  
  "Turn off the light" - entity: light
  "Switch on the TV" - entity: TV

• Intents: The purpose, or goal, expressed in a user's utterance.

  Example:
  
  "Turn off the light" - purpose: off a device
  "Switch on the TV" - purpose: on a device

1. Authoring the model: by defining entities, intents, and utterances.
2. Publish the model: For intent and entity prediction based on user input.

1. Machine-Learned
2. List
3. RegEx
4. Pattern.any

1. Language Service:
Custom question answering feature that enables you to create a knowledge base of question and answer pairs that can be queried using natural language input.

2. Azure Bot Service: Framework for developing, publishing, and managing bots on Azure.

1. The knowledge base ID
2. The knowledge base endpoint
3. The knowledge base authorization key