What unsupervised learning is

Unsupervised learning works with features only (X) — no labels yy. As Yann LeCun put it, if intelligence were a cake, unsupervised learning would be the cake itself, and supervised learning would just be the icing.

Goals include:

• clustering: group similar points
• dimensionality reduction: compress data while keeping its structure
• anomaly detection: find rare/unusual samples
• association rules: discover co-occurrence patterns

This phase mirrors Chapters 8 and 9 of Hands-On Machine Learning (Géron): first the clustering and anomaly-detection algorithms that group and score unlabeled instances, then the projection and manifold-learning techniques — above all Principal Component Analysis (PCA) — that fight the curse of dimensionality by compressing high-dimensional data down to something a model (or a human) can actually work with.

  flowchart LR
  X["Unlabeled data (X)"] --> A[Unsupervised algorithm]
  A --> C[Clustering]
  A --> D["Dimensionality Reduction (PCA)"]
  A --> AN[Anomaly Detection]
  A --> R[Association Rules]

Phase 6 topics

1. Introduction to Clustering
2. K-Means Clustering Algorithm
3. Hierarchical Clustering (Dendrograms)
4. DBSCAN: Density-Based Clustering
5. Anomaly Detection with Isolation Forests
6. Association Rule Learning (Apriori Algorithm)
7. Principal Component Analysis (PCA)
8. t-SNE and Manifold Learning

Common real-world uses

• customer segmentation
• grouping products by similarity
• detecting fraud/anomalies
• analyzing basket/cart patterns
• compressing high-dimensional data (e.g. MNIST) for faster training and 2D/3D visualization