ai900 basics 1

 

What is AI?>

  • Definition: AI is the simulation of human intelligence in machines so they can think, learn, and make decisions.

  • Goal: To create systems that can perform tasks normally requiring human intelligence (e.g., problem solving, learning, reasoning).

Differences between AI, ML, and DL

  1. AI: Broad field – machines that can act smart.
    Example: Chatbots, recommendation systems.

  2. ML: Subset of AI – machines learn patterns from data without being explicitly programmed.
    Example: Spam email detection.

  3. DL (Deep Learning): Subset of ML – uses neural networks with many layers to process complex data.
    Example: Image recognition in self-driving cars.

Exam Keypoint: AI ⊃ ML ⊃ DL (AI is the biggest set).

Common Applications of AI in Daily Life

  • Voice assistants (Siri, Alexa, Google Assistant).

  • Recommendations (Netflix, Amazon).

  • Fraud detection (banking).

  • Autonomous vehicles.

  • Smart home devices (thermostats, security cameras).

Types of AI

1. Narrow AI (Weak AI)

  • Specialized in one task only.

  • Example: Face recognition on phone.

2. General AI (Strong AI)

  • Hypothetical – can perform any intellectual task like a human.

  • Still under research, not yet achieved.

Levels of AI

  1. Reactive Machines – respond only to current input, no memory.
    Example: Chess program (IBM Deep Blue).

  2. Limited Memory – can use past data for decisions.
    Example: Self-driving cars.

  3. Theory of Mind – future AI that understands human emotions/intent.
    Example: Not yet available.

  4. Self-Aware AI – hypothetical AI with consciousness.
    Example: Does not exist yet.

Key Domains of AI

  1. Machine Learning (ML)

    • Teaches machines to learn from data.

    • Example: Predicting movie ratings based on user history.

  2. Natural Language Processing (NLP)

    • AI understands, interprets, and generates human language.

    • Example: Chatbots, language translation (Google Translate).

  3. Computer Vision

    • AI interprets and analyzes images or videos.

    • Example: Face recognition in Facebook photos.

  4. Conversational AI

    • AI systems that can interact through voice or text.

    • Example: Virtual assistants, customer service chatbots.

✅ AI-900 Key Quick Notes

  • AI = Simulation of human intelligence.

  • AI > ML > DL.

  • Types: Narrow (real) vs General (future).

  • Levels: Reactive, Limited Memory, Theory of Mind, Self-Aware.

  • Domains: ML, NLP, Computer Vision, Conversational AI.

  • Applications: Voice assistants, recommendations, fraud detection, self-driving cars.

PART 2: Machine Learning (ML) Basics

What is Machine Learning?

  • Definition: ML is when machines learn patterns from data and improve performance without being explicitly programmed.

  • Example: Email spam filter learns to classify spam vs non-spam from past labeled emails.

How ML differs from traditional programming

  • Traditional programming: Rules + Data → Output.

  • Machine Learning: Data + Output (labels) → Algorithm learns rules.

Types of Machine Learning

  1. Supervised Learning

    • Learn from labeled data (input + correct output).

    • Example: Predict house price (input: size, location; output: price).

  2. Unsupervised Learning

    • Learn patterns from unlabeled data.

    • Example: Group customers into clusters based on purchase behavior.

  3. Reinforcement Learning

    • Agent learns by trial and error with rewards/punishments.

    • Example: Game AI learning to play chess.

Common ML Algorithms (basic)

  1. Classification – Predict categories.

    • Example: Email → Spam / Not Spam.

    • Algorithms: Decision Trees, Logistic Regression.

  2. Regression – Predict continuous values.

    • Example: Predict stock price.

  3. Clustering – Group data without labels.

    • Example: Market segmentation (K-Means groups customers).

ML Model Lifecycle

  1. Data Collection – Gather data (images, text, numbers).

  2. Data Preprocessing – Clean, normalize, handle missing values.

  3. Model Training – Feed data to ML algorithm to learn patterns.

  4. Model Evaluation – Test accuracy/performance on unseen data.

  5. Deployment – Use model in real applications (apps, services).

Evaluation Metrics

  1. Accuracy – % of correct predictions.

    • Works well when classes are balanced.

  2. Precision – Of predicted positives, how many are correct.

    • Example: Of emails marked spam, how many truly spam.

  3. Recall – Of actual positives, how many were correctly predicted.

    • Example: How many spam emails were caught.

  4. F1 Score – Balance between precision and recall.

    • Useful when data is imbalanced.

  5. Confusion Matrix – Table showing:

    • True Positives, False Positives, True Negatives, False Negatives.

✅ AI-900 Quick Notes

  • ML = Learn from data, not rules.

  • Types: Supervised (labeled), Unsupervised (unlabeled), Reinforcement (rewards).

  • Algorithms: Classification (Decision Tree, Logistic Regression), Regression, Clustering (K-Means).

  • Lifecycle: Collect → Preprocess → Train → Evaluate → Deploy.

  • Metrics: Accuracy, Precision, Recall, F1, Confusion Matrix.

Computer Vision Basics

What is Computer Vision?

  • Computer Vision = AI that makes computers "see" and understand images/videos.

Key point: It extracts information (objects, text, faces) from images.

Real-life Applications

  • Facial recognition → unlock phone, security.

  • Object detection → self-driving cars.

  • Medical imaging → detect diseases.

Common Computer Vision Tasks

  1. Image Classification → Identify one object in an image.
    Example: "Is this a cat or dog?"

  2. Object Detection → Find multiple objects + location in image.
    Example: Detect cars, people in traffic photo.

  3. Semantic Segmentation → Label each pixel.
    Example: Separate road, car, pedestrian in an image.

  4. OCR (Optical Character Recognition) → Extract text from images.
    Example: Scan printed receipts.

Exam tip: Classification = one label, Detection = multiple + location, Segmentation = pixel-level, OCR = text.

Azure Services for Computer Vision

  1. Azure Computer Vision API

    • Prebuilt models.

    • Tasks: OCR, tagging images, analyzing content.

    • Easy, no training needed.

  2. Azure Custom Vision

    • Train your own model with your data.

    • Useful for specific needs (e.g., detect brand logos).

Natural Language Processing (NLP) Basics

What is NLP?

  • NLP = AI that makes computers understand and process human language (text or speech).

Key point: It bridges human language and machine understanding.

Everyday Uses

  • Chatbots → customer support.

  • Translators → Google Translate, Azure Translator.

  • Sentiment Analysis → detect emotions (positive/negative).

Common NLP Tasks

  1. Text Classification → Assign category to text.
    Example: Spam vs Non-spam emails.

  2. Named Entity Recognition (NER) → Find names, dates, places in text.
    Example: "Bill Gates founded Microsoft in 1975" → Bill Gates = person, Microsoft = organization, 1975 = date.

  3. Language Translation → Convert text between languages.
    Example: English → French.

  4. Sentiment Analysis → Detect opinion/emotion.
    Example: "The movie was amazing" → Positive.

Exam tip: NER = identify entities, Sentiment = opinion, Classification = category.

Azure Services for NLP

  1. Azure Text Analytics

    • Prebuilt API.

    • Tasks: sentiment, key phrases, language detection, NER.

  2. Azure Translator

    • Translate text in real time.

    • Supports multiple languages.

  3. Azure LUIS (Language Understanding)

    • Custom NLP for intent recognition.

    • Example: In a chatbot, "Book me a flight" → intent = booking.

Using Microsoft Azure for Building and Deploying AI Solutions

Azure AI Services Overview

  • Cognitive Services → Ready-made APIs for vision, speech, language, decision-making. No training needed.

  • Azure Machine Learning (Azure ML) → Platform to train, deploy, and manage ML models.

  • Azure AI Studio → Central place to explore, build, and deploy AI solutions.

Responsible AI Principles (Microsoft)

  1. Fairness → AI should not be biased.

  2. Reliability → AI must work safely and consistently.

  3. Privacy & Security → Protect user data.

  4. Inclusiveness → Accessible to all people.

  5. Transparency → Make AI decisions understandable.

  6. Accountability → Humans responsible for AI outcomes.

Exam tip: Memorize these 6 core principles.

Low-code / No-code AI Solutions

  • Azure ML Studio (Designer) → Drag-and-drop tool for building ML models.

  • AutoML → Automates model selection and training with minimal coding.

PART 7: Responsible AI and Ethics

Bias in AI

  • Problem: AI may treat groups unfairly if trained on biased data.

  • Example: Hiring AI preferring one gender.

Explainability

  • AI decisions must be understandable by humans.

  • Builds trust.

Privacy Concerns

  • Protect sensitive data (e.g., medical records).

  • Use secure storage and compliance rules.

Sustainability of AI Systems

  • AI should be energy-efficient and environmentally friendly.

  • Example: Optimize training to reduce carbon impact.

Quick Exam Pointers

  • Cognitive Services = ready-to-use.

  • Azure ML = custom models.

  • AI Studio = central workspace.

  • Responsible AI = Fairness, Reliability, Privacy, Inclusiveness, Transparency, Accountability.

  • AutoML/ML Studio = less coding.

  • Ethics focus = Bias, Explainability, Privacy, Sustainability.


QUICK REVISE

AI-900 Quick Revision Sheet

PART 1: AI Fundamentals

  • AI → Machines simulate human intelligence.

  • ML vs Traditional → ML learns from data, traditional uses rules.

  • Types of ML →

    • Supervised (labeled)

    • Unsupervised (no labels)

    • Reinforcement (rewards/punishments)

  • Common ML tasks → Classification, Regression, Clustering.

PART 2: ML Lifecycle

  • Data collection → Preprocessing → Training → Evaluation → Deployment.

  • Important: More data = better model.

PART 3: Computer Vision

  • Computer Vision → AI understands images/videos.

  • Tasks →

    • Classification (one label)

    • Object detection (multiple objects + location)

    • Segmentation (pixel-level labels)

    • OCR (text from images)

  • Azure → Computer Vision API (prebuilt), Custom Vision (train your own).

PART 4: NLP (Natural Language Processing)

  • NLP → AI that understands human language.

  • Tasks →

    • Text classification (spam detection)

    • NER (find names, dates, places)

    • Translation

    • Sentiment analysis

  • Azure → Text Analytics, Translator, LUIS.

PART 5: Conversational AI

  • Conversational AI → Chatbots (text) + Voice assistants (speech).

  • Azure →

    • Bot Service (build/deploy bots)

    • QnA Maker (FAQ → bot)

    • Integrate with Teams, web apps.

PART 6: Azure AI Solutions

  • Cognitive Services → Ready APIs (vision, speech, language).

  • Azure ML → Train, deploy, manage ML models.

  • AI Studio → Central workspace for AI.

  • Low-code/No-code → ML Studio Designer (drag-drop), AutoML (automated training).

PART 7: Responsible AI & Ethics

  • Principles → Fairness, Reliability, Privacy & Security, Inclusiveness, Transparency, Accountability.

  • Bias → Avoid unfair treatment.

  • Explainability → Humans must understand AI decisions.

  • Privacy → Protect sensitive data.

  • Sustainability → Build energy-efficient AI systems.

EXAM TIPS

  • Know the difference between Cognitive Services vs Custom Models (Azure ML).

  • Memorize the 6 Responsible AI principles.

  • Match each task with its Azure service.

  • Remember low-code tools = ML Studio & AutoML.


AI-900 Exam Tips

PART 1: Fundamentals of AI

  • Definition: AI = simulate human intelligence.

  • AI vs ML vs DL:

    • AI = broad concept.

    • ML = subset (learn from data).

    • DL = subset of ML (neural networks).

  • Applications: chatbots, self-driving cars, recommendation systems.

  • Types of AI:

    • Narrow AI = task-specific (most current AI).

    • General AI = human-level (future).

  • Categories:

    • Reactive Machines → only respond.

    • Limited Memory → use past data (self-driving cars).

    • Theory of Mind → understand human emotions (not yet real).

    • Self-aware → future.

  • Domains: ML, NLP, Computer Vision, Conversational AI.

Tip: Expect questions comparing AI vs ML vs DL, or Narrow vs General AI.

PART 2: Machine Learning (ML) Basics

  • ML = learn patterns from data.

  • Traditional vs ML → Traditional = rules, ML = data-driven.

  • Types:

    • Supervised = labeled data.

    • Unsupervised = no labels.

    • Reinforcement = rewards.

  • Algorithms: Classification (spam filter), Regression (house price), Clustering (customer groups).

  • Lifecycle: Collect → Clean → Train → Evaluate → Deploy.

  • Evaluation Metrics:

    • Accuracy = correct predictions %

    • Precision = how many predicted positives are true

    • Recall = how many actual positives found

    • F1 = balance of precision & recall

    • Confusion Matrix = table of predictions

Tip: Focus on Supervised vs Unsupervised and metrics differences.

PART 3: Computer Vision Basics

  • Definition: AI that "sees" images/videos.

  • Tasks:

    • Classification = one object

    • Object Detection = multiple + location

    • Segmentation = pixel-level labels

    • OCR = extract text

  • Azure Services:

    • Computer Vision API = prebuilt (tags, OCR).

    • Custom Vision = train with your own images.

Tip: Know difference → API = ready-to-use, Custom Vision = train your own.

PART 4: NLP Basics

  • Definition: AI that understands text/speech.

  • Tasks:

    • Text classification = spam detection

    • NER = find names, dates, orgs

    • Translation = convert languages

    • Sentiment = opinion detection

  • Azure Services:

    • Text Analytics = sentiment, key phrases, NER.

    • Translator = real-time translation.

    • LUIS = intent recognition in chatbots.

Tip: Match NLP task → Azure service correctly.

PART 5: Conversational AI

  • Definition: AI that talks to users.

  • Chatbots = text-based.

  • Voice assistants = speech-based.

  • Azure Services:

    • Bot Service = build bots.

    • QnA Maker = FAQ → bot.

    • Integrate into Teams, websites.

Tip: QnA Maker = FAQ bots, Bot Service = general bots.

PART 6: AI on Azure

  • Cognitive Services = ready-made APIs (vision, speech, language).

  • Azure ML = build, train, deploy custom ML.

  • AI Studio = central place for AI tools.

  • Responsible AI Principles:

    • Fairness, Reliability, Privacy/Security, Inclusiveness, Transparency, Accountability.

  • Low/No Code:

    • ML Studio Designer = drag-and-drop ML.

    • AutoML = automatic model building.

Tip: Memorize 6 Responsible AI Principles (they are asked directly).

PART 7: Responsible AI and Ethics

  • Bias = unfair predictions due to bad data.

  • Explainability = humans must understand AI decisions.

  • Privacy = protect sensitive data.

  • Sustainability = efficient AI to save energy.

Tip: Questions often test your ability to identify risks in AI ethics (bias, privacy, explainability).

Final Exam Strategy

  1. Focus on differences (AI vs ML vs DL, Supervised vs Unsupervised, Classification vs Detection).

  2. Know which Azure service solves which task.

  3. Memorize Responsible AI principles.

  4. Expect scenario-based questions (e.g., “Which service to use for extracting text from receipts?” → OCR).

AI-900 Exam Hints

PART 1: Fundamentals of AI

  • AI = umbrella term, ML and DL are subsets.

  • AI vs ML vs DL → remember:

    • AI = simulating intelligence

    • ML = learning from data

    • DL = neural networks

  • Applications → self-driving cars, recommendation systems, chatbots.

  • Types of AI:

    • Narrow AI = specific tasks (today’s AI).

    • General AI = human-like intelligence (future).

  • Categories (Reactive → Limited Memory → Theory of Mind → Self-Aware) → only first 2 exist today.

  • Key domains always = ML, NLP, CV, Conversational AI.

Hint: If the question asks “Which AI is used today?” → Narrow AI.

PART 2: Machine Learning Basics

  • Traditional vs ML → Traditional = hard-coded rules, ML = learns from data.

  • Types of ML:

    • Supervised → labeled data, prediction.

    • Unsupervised → unlabeled, patterns.

    • Reinforcement → trial & error + reward.

  • Algorithms:

    • Classification = categories (spam filter).

    • Regression = continuous value (house price).

    • Clustering = grouping (customer segmentation).

  • ML Lifecycle → Data → Preprocessing → Training → Evaluation → Deployment.

  • Metrics:

    • Accuracy = overall correctness.

    • Precision = correct positives.

    • Recall = captured actual positives.

    • F1 = balance between precision & recall.

    • Confusion Matrix = summary table.

Hint: If question has “labeled data” → answer = Supervised learning.

PART 3: Computer Vision Basics

  • Computer Vision = making computers see and analyze images.

  • Tasks:

    • Classification = one object.

    • Detection = multiple objects + positions.

    • Segmentation = pixel-level labeling.

    • OCR = extract text.

  • Azure Services:

    • Computer Vision API = prebuilt.

    • Custom Vision = train on your own dataset.

Hint: Text extraction from scanned receipts → OCR (Computer Vision API).

PART 4: NLP Basics

  • NLP = AI that understands human language (text/speech).

  • Tasks:

    • Text Classification = spam detection.

    • NER = find entities (names, dates, places).

    • Translation = convert languages.

    • Sentiment Analysis = detect opinions.

  • Azure Services:

    • Text Analytics = sentiment, key phrases, NER.

    • Translator = real-time translation.

    • LUIS = intent recognition in chatbots.

Hint: Intent detection in chatbot → LUIS.

PART 5: Conversational AI

  • Conversational AI = AI that interacts with users.

  • Chatbots = text-based.

  • Voice Assistants = speech-based.

  • Azure Services:

    • Bot Service = build bots.

    • QnA Maker = FAQ → chatbot.

    • Integration with Teams/web.

Hint: FAQ document → QnA Maker.

PART 6: AI on Azure

  • Cognitive Services = ready-to-use APIs.

  • Azure ML = build, train, deploy custom ML.

  • AI Studio = central place for AI services.

  • Responsible AI Principles (must memorize):

    • Fairness, Reliability, Privacy & Security, Inclusiveness, Transparency, Accountability.

  • Low/No-code:

    • ML Studio Designer = drag-and-drop.

    • AutoML = automated training/model selection.

Hint: If no coding required → Designer or AutoML.

PART 7: Responsible AI & Ethics

  • Bias = unfair outcomes due to poor training data.

  • Explainability = humans must understand why AI decided something.

  • Privacy = protect sensitive data.

  • Sustainability = energy-efficient AI.

Hint: If question mentions trust or transparency → choose Explainability.

Final Quick Hints

  1. Always map problem → Azure service.

  2. Remember Narrow AI = today, General AI = future.

  3. Supervised = labels, Unsupervised = patterns, Reinforcement = rewards.

  4. OCR = text extraction.

  5. LUIS = intent recognition, Translator = language conversion.

  6. Responsible AI = 6 principles (memorize!).