0.2 Types of Machine Learning★★★

# Three Ways to Learn

• A child learns fruit names when a parent points and says "apple".
• Later, she sorts toys into groups that nobody named for her.
• And she learns to ride a bicycle by falling and trying again.
• Machines learn in the same three ways.

# Types of Machine Learning PYQ 6M

flowchart TD
    A["<b>Data</b>"]:::blue --> B{"Labelled<br>examples?"}:::amber
    B -->|"Yes"| C["<b>Supervised</b>"]:::green
    B -->|"No"| D{"Learns from<br>rewards?"}:::amber
    D -->|"Yes"| E["<b>Reinforcement</b>"]:::green
    D -->|"No"| F["<b>Unsupervised</b>"]:::green
    classDef blue fill:#5b84d424,stroke:#5b84d480
    classDef amber fill:#d6a24424,stroke:#d6a24480
    classDef red fill:#d0707024,stroke:#d0707080
    classDef green fill:#52a86e24,stroke:#52a86e80

# Supervised Learning

Supervised learning is learning from labelled data: each example has an input x and the correct output y, and the model learns the mapping x → y.

• Like a student learning with an answer key.

Two kinds (→ 1.1)

1. Classification: output is a category. Example: spam / not spam; disease yes / no.
2. Regression: output is a number. Example: house price; tomorrow's temperature.

Algorithms: linear and logistic regression, KNN, decision trees, Naive Bayes, SVM.

# Unsupervised Learning PYQ 1M

Unsupervised learning is learning from unlabelled data, where the goal is to find hidden structure, patterns or groupings.

• No answer key; the model discovers structure on its own.

Main tasks

1. Clustering: group similar items. Example: customer segmentation (→ 2.1).
2. Dimensionality reduction: compress many features into few (→ 2.4).
3. Association rules: find items that occur together. Example: "bread → butter" in shopping baskets.

Algorithms: K-means, hierarchical clustering, PCA, Apriori.

# Reinforcement Learning

Reinforcement learning (RL) is learning by trial and error: an agent acts in an environment and learns from rewards and penalties to maximise total reward (→ 5.4).

• Examples: game-playing AI (AlphaGo), robot navigation, self-driving decisions.

# Other Types

• Semi-supervised learning: a few labelled examples plus many unlabelled ones (→ 5.2).
  • Example: a photo app labels a few faces you tag, then groups the rest.
• Active learning: the model asks a human to label the most useful examples (→ 5.3).
• Online learning: the model updates continuously as new data arrives (→ 5.1).
  • Example: a stock-price model updated every minute.

# Supervised vs Unsupervised vs Reinforcement

# Matching Applications to Types PYQ 1M

# Choosing the Type: A Movie-Review Dataset

• Situation: historical movie reviews, each with a class label (positive / negative).
• Choice: supervised learning, classification (sentiment analysis).
• Why:
  1. Every review already has the correct label, so the model can learn from the answer key.
  2. The output is a category, so it is classification, not regression.
  3. Suitable algorithms: Naive Bayes (→ 1.7), logistic regression, SVM.
• Without labels, the same reviews could only be grouped by clustering (unsupervised).