Unsupervised Learning
Unsupervised learning is a machine learning approach in which a model finds patterns, structure, or groupings in data without being given labeled outputs.
Definition
Unsupervised learning is a machine learning approach in which a model finds patterns, structure, or groupings in data without being given labeled outputs.
Overview
Unlike supervised learning, unsupervised learning works with unlabeled data — there is no "correct answer" provided during training. Instead, the model's job is to discover inherent structure in the data on its own, such as grouping similar items together (clustering), reducing the number of variables while preserving important information (dimensionality reduction), or identifying which data points deviate from normal patterns (anomaly detection). Because it does not require labeled data, unsupervised learning is particularly useful when labels are expensive, impractical, or simply unavailable — which is common in real-world datasets. Common algorithms include k-means clustering for grouping, and principal component analysis (PCA) for dimensionality reduction, though many modern approaches use neural networks as well. Unsupervised learning is closely related to, but distinct from, self-supervised learning, a more recent paradigm that generates its own supervisory signal from unlabeled data — a technique that underlies the pretraining of most large language models and other foundation models in use today.
Key Concepts
- Learns patterns from unlabeled data without predefined correct answers
- Commonly used for clustering, dimensionality reduction, and anomaly detection
- Does not require expensive or time-consuming manual labeling
- Useful for exploratory data analysis and discovering hidden structure
- Includes classic algorithms like k-means clustering and PCA
- Distinct from but related to self-supervised learning