Ksama Arora

Microsoft Certified: Azure AI Fundamentals (AI-900)

Module 1 - Fundamental AI Concepts

AI is a software that imitates human behaviours and capabilities.(MNCDKG)

6 principles: Fairness, Reliability and Safety, Privacy and Security, Inclusivness, Transparency, Accountability.

Module 2 - Use Automated Machine Learning in Azure Machine Learning

Machine Learning is an AI tool that uses mathematics and statistics to create a model that can predict unknown values. F(x)=y where x -> features and y -> labels we are trying to predict.

1) Supervised Machine Learning: Algo trained on labelled dataset (labels & features known)

• Regression - used to predict a numerical value - does not rely on randomly generated data for training

• Classification - used to predict a category or class

2) Unsupervised Machine Learning: Algo trained without any predefined labels (features and labels not known)

• Clustering - grouping similar items together based on their features without relying on training and validating label predictions

Module 3 - Create a Regression Model with Azure Machine Learning Designer

Azure Machine Learning Studio:

To begin use of it, you need to associate resource created in Azure portal to Azure Machine Learning studio.

Azure Automated ML: automatically trains ML model by providing dataset and picking label i.e. automatically pick algo - ((DICY))allows to include custom python script - uses solns without need of programming.

Azure ML Designer: provides drag-and-drop interface/ canvas to train, test and deploy ML models - enables to save progress as a pipeline draft - allows custom python & R fns and not JS functions.

Module 4 - Create a classification model with Azure Machine Learning Designer

Refer to Lab 3 link Explore classification with Azure Machine Learning Designer

UNDERSTAND STEPS FOR CLASSIFICATION:

1) Prepare Data, 2) Train Model

3) Evaluate Performance:

Confusion Matrix: It is a tool to assess the quality of a classification models predictions.

Metrics that can be derived from confusion matrix: Accuracy, Precision, Recall, F1 Score.

Module 5 - Create a clustering model with Azure ML Designer

Refer to Lab 4 link Create a clustering model with Azure ML Designer

UNDERSTAND STEPS FOR CLUSTERING:

1) Prepare data

2) Train model: To train a clustering model, you need to apply a clustering algorithm to the data. K- Means Clustering Algorithm groups items into numbers of clusters, or centroids i.e. ‘K’.

3) Evaluate Performance: Metrics are Average Distance to Other Center, Average Distance to Cluster Center, Number of Points, Maximal Distance to Cluster Center, Silhouette.

5) Deployment: (explained earlier)

Module 6 - Computer Vision

Computer Vision - manipulation and analysis of pixel values in images.

Azure Services for computer vision:

• Azure Computer Vision Read API: Describe, tag, (brand, object, face) detect, categorize img, landmark and celeb, generate thumbnails, and detect inappropriate content
• Azure Custom Vision: Analayze, classify images and detect objects (class label, probability and bounding box)
• Azure AI Face
• Azure AI Video
• Azure AI Document Intelligence / Azure Forms Recognizer
• Azure Cognitive Services (includes all services - uses only one endpoint and key - can be used for both training and prediction)

Fundamentals of Facial Recognition

To detect faces in Vision Studio, first make face resource in Azure AI Vision and the upload and analyze photos.

Azure Services:

• Azure AI Vision: face detection & analysis (bounding box coordinates)
• Azure AI Video Indexer: detect and identify faces in video
• Azure AI Face: detect, recognize, and analyze faces - returns set of coordinates for each face, a rect bounding box
  • Face identification: One-to-many face matching
  • Face verification: One-to-one face matching
• Azure Cognitive Services: The one service to rule them all!

Fundamentals of Optical Character Recognition

Azure Services:

• Azure AI Vision - to track utilization and costs for AI Vision resource separately - for image analysis
• Azure AI Language - for text analytics
• Azure AI Cognitive Services - include Azure AI Vision and many other services - used when want to simplify administration and deployment

Fundamentals of Azure AI Document Intelligence

Azure AI Document Intelligence/ Azure Forms Recognizer service: best way to read text from images of reciepts, invoices and forms, automates process of extracting, understanding and saving data in text

• Document Analysis: ability to extract text, layout, and key-value pairs - location of text along with bounding box coordinates

Module 7 - Natural Language Processing

Process to make bot:

• First make Language resource - use Language Studio’s custom qs ans feature - qs ans can be generated by importing an existing FAQ doc/ web page or entered manually with additional manual entries - save it.
• After created and deployed a knowledge base, you can deliver it to users through a bot - create custom bot using Microsoft Bot Framework SDK ( writes code - integrates with knowledge base)

NLP Services:

• Text (Azure AI Language): includes lang detection, sentiment analysis, entity recognition, entity linking, extracting key phrases, summarization, personal identifying information

• Speech Detection
• Translator - only text-to-text translation
• Language Understanding: interpret and implement commands
• Azure Cognitive Services

• Azure Cognitive Services: 1 service to rule them all

Fundamentals of Conversational Language Understanding

3 Core Concepts: Utterances, Entities and Intent.

Azure Services

• Azure AI Language - under which LUIS (Language Understanding Service) works - standalone service that offers authoring, training and prediction

• Azure Cognitive Service - 1 service to rule them all! (NOTE: This can only be used for predictions and not for authoring models)

Conversational AI: AI systems that can chat with humans e.g. bot answering, automated responses Note: Conversational AI is only used for conversing, and is not same as NLP

Two core services: Azure AI Language and Azure AI Bot Service.