Domain 1: Fundamentals of AI and ML (20%)

← Overview · Next Domain →

---

1.1: AI, ML, and DL Hierarchy

┌─────────────────────────────────────┐
│    Artificial Intelligence (AI)     │  Broadest: Machines that mimic human intelligence
│  ┌───────────────────────────────┐  │
│  │   Machine Learning (ML)       │  │  Subset: Learn from data without explicit programming
│  │  ┌─────────────────────────┐  │  │
│  │  │  Deep Learning (DL)     │  │  │  Subset: Neural networks with multiple layers
│  │  └─────────────────────────┘  │  │
│  └───────────────────────────────┘  │
└─────────────────────────────────────┘

Why This Matters

Exam questions test whether you understand the hierarchy: All DL is ML, all ML is AI, but not all AI is ML.

---

Types of Machine Learning

ML Learning Types

1 / 3

❓

What type of ML uses labeled data?

(Click to reveal)

💡

Supervised Learning
Uses features (X) and labels (y) to learn a mapping. Examples: spam detection, house price prediction.

1. Supervised Learning

Definition: Learn from labeled data

Components:

• Features (X): Input variables
• Labels (y): Known output/target
• Model: Learns mapping from X to y

Types:

• Classification: Predict categories (spam/not spam)
• Regression: Predict continuous values (house price)

Examples:

• Email spam detection
• Image classification (cat vs dog)
• House price prediction
• Customer churn prediction

2. Unsupervised Learning

Definition: Learn patterns from unlabeled data

Types:

• Clustering: Group similar items
• Dimensionality Reduction: Reduce features while preserving information
• Anomaly Detection: Find outliers

Examples:

• Customer segmentation
• Anomaly detection in network traffic
• Product recommendations

3. Reinforcement Learning

Definition: Learn through trial and error with rewards

Components:

• Agent: Learns and makes decisions
• Environment: What agent interacts with
• Actions: What agent can do
• Rewards: Feedback for actions

Examples:

• Game playing (AlphaGo)
• Robotics
• Autonomous vehicles
• Resource optimization

Learning Type	Data	Goal	Example
Supervised	Labeled	Predict labels	Spam detection
Unsupervised	Unlabeled	Find patterns	Customer groups
Reinforcement	Rewards/penalties	Maximize rewards	Game AI

---

ML Model Performance Metrics

Classification Metrics

Confusion Matrix:

                  Predicted
              Positive  Negative
Actual  Pos      TP        FN
        Neg      FP        TN

Classification Metrics

1 / 3

❓

When should you prioritize Precision?

(Click to reveal)

💡

When false positives are costly
Precision = TP / (TP + FP)
Example: Spam filter (do not mark important emails as spam)

Key Metrics:

• Accuracy = (TP + TN) / Total

  • Good when classes are balanced

• Precision = TP / (TP + FP)

  • "Of all predicted positives, how many are actually positive?"
  • Use when false positives are costly

• Recall = TP / (TP + FN)

  • "Of all actual positives, how many did we catch?"
  • Use when false negatives are costly

Exam Scenario

Medical diagnosis: Prefer high recall (catch all diseases, even with false positives) Spam filter: Prefer high precision (don't mark important emails as spam)

Regression Metrics

• MAE (Mean Absolute Error): Average absolute difference
• MSE (Mean Squared Error): Average squared difference
• RMSE (Root Mean Squared Error): Square root of MSE

---

Overfitting vs. Underfitting

Model Fit Issues

1 / 3

❓

What is Underfitting?

(Click to reveal)

💡

Model too simple
Poor on BOTH training and test data.
Solution: More complex model, more features.

Issue	Description	Performance	Solution
Underfitting	Model too simple	Poor on training AND test	More complex model, more features
Good Fit	Just right	Good on both	✅ This is the goal
Overfitting	Memorized training data	Great on training, poor on test	More data, regularization, simpler model

Visual:

Underfitting:  🙁 Training: 70%  🙁 Test: 65%
Perfect Fit:   ✅ Training: 92%  ✅ Test: 90%
Overfitting:   ✅ Training: 99%  🙁 Test: 72%

---

ML Development Lifecycle

1. Business Problem Definition

   • What are we trying to predict/optimize?
   • What's the success metric?

2. Data Collection and Preparation

   • Gather data
   • Clean data (handle missing values, outliers)
   • Split data (train/validation/test)

3. Feature Engineering

   • Create useful features
   • Transform variables
   • Encode categorical variables

4. Model Training

   • Choose algorithm
   • Train on training data
   • Tune hyperparameters

5. Model Evaluation

   • Test on validation data
   • Check metrics (accuracy, precision, recall)
   • Compare multiple models

6. Model Deployment

   • Deploy to production
   • Create API endpoint
   • Monitor performance

7. Model Monitoring and Maintenance

   • Track model drift
   • Retrain when needed
   • Update as data changes

Exam Focus

The exam tests whether you understand that ML is iterative, not one-and-done. Models need monitoring and retraining.

---

← Overview · Next Domain →